WEBVTT 00:00:00.000 --> 00:00:03.900 align:middle line:84% MARCO RAMONI: Today I'm going to talk to you about the-- 00:00:03.900 --> 00:00:07.260 align:middle line:84% about basic genetics, what geneticists do, 00:00:07.260 --> 00:00:11.970 align:middle line:84% and how genetics is moving into the genomic area 00:00:11.970 --> 00:00:18.750 align:middle line:84% by increasing the size, the scope, and the quality 00:00:18.750 --> 00:00:20.910 align:middle line:90% of genetic studies. 00:00:20.910 --> 00:00:22.785 align:middle line:90% We'll do-- 00:00:22.785 --> 00:00:23.410 align:middle line:90% [DOOR CREAKING] 00:00:23.410 --> 00:00:25.840 align:middle line:90% You can close the door. 00:00:25.840 --> 00:00:30.870 align:middle line:84% So the origin of all of this is-- 00:00:30.870 --> 00:00:33.810 align:middle line:84% I will put myself at the particular time point, 00:00:33.810 --> 00:00:36.780 align:middle line:84% and the origin of all of this is the publication 00:00:36.780 --> 00:00:38.940 align:middle line:84% of the Human-- of the results of the Human Genome 00:00:38.940 --> 00:00:42.330 align:middle line:84% Project, a first draft of the human genome. 00:00:42.330 --> 00:00:45.060 align:middle line:84% And when they announced the publication 00:00:45.060 --> 00:00:50.850 align:middle line:84% of the human genome, the first item on the future agenda 00:00:50.850 --> 00:00:55.080 align:middle line:84% was a SNP map, and they said, this SNP map 00:00:55.080 --> 00:00:57.390 align:middle line:84% promises to revolutionize both mapping diseases 00:00:57.390 --> 00:00:59.230 align:middle line:90% and tracking human history. 00:00:59.230 --> 00:01:03.240 align:middle line:84% And I will try to show you how you do both using SNPs, 00:01:03.240 --> 00:01:06.010 align:middle line:84% and if you don't know what SNPs are, hold onto your breath. 00:01:06.010 --> 00:01:07.494 align:middle line:90% You will know it in a second. 00:01:07.494 --> 00:01:09.025 align:middle line:90% [DOOR CREAKING] 00:01:09.025 --> 00:01:09.525 align:middle line:90% Please. 00:01:09.525 --> 00:01:12.130 align:middle line:90% 00:01:12.130 --> 00:01:17.550 align:middle line:84% The SNPs, I will tell you in a second in more-- in detail. 00:01:17.550 --> 00:01:21.480 align:middle line:84% SNPs are things called single nucleotide polymorphisms, 00:01:21.480 --> 00:01:25.500 align:middle line:84% and there are subtle variations in the human genome 00:01:25.500 --> 00:01:26.760 align:middle line:90% across individuals. 00:01:26.760 --> 00:01:27.870 align:middle line:90% What this mean? 00:01:27.870 --> 00:01:30.600 align:middle line:84% It means that one of the most startling statistics out 00:01:30.600 --> 00:01:33.840 align:middle line:84% of the Genome Project is that we're all similar, very, very 00:01:33.840 --> 00:01:35.280 align:middle line:90% similar at genetic level. 00:01:35.280 --> 00:01:38.220 align:middle line:84% But there are some very, very subtle variations, 00:01:38.220 --> 00:01:42.870 align:middle line:84% one base pair swapped here and there, that make us different 00:01:42.870 --> 00:01:45.890 align:middle line:84% and, most of all, makes our genomes different. 00:01:45.890 --> 00:01:51.690 align:middle line:84% And what you can say is that this is the real meat 00:01:51.690 --> 00:01:52.980 align:middle line:90% of the human genome. 00:01:52.980 --> 00:01:55.020 align:middle line:84% People paid billions of dollars not 00:01:55.020 --> 00:01:57.480 align:middle line:84% for the curiosity of a handful of scientists, 00:01:57.480 --> 00:02:00.390 align:middle line:84% but because the promise was that, once we know 00:02:00.390 --> 00:02:02.460 align:middle line:84% the human genome, it would be easier 00:02:02.460 --> 00:02:05.670 align:middle line:84% to track down the genetic causes of diseases 00:02:05.670 --> 00:02:09.539 align:middle line:84% and the genetic causes of-- and the-- possibly 00:02:09.539 --> 00:02:12.690 align:middle line:84% find some cure for them and get a better understanding. 00:02:12.690 --> 00:02:16.890 align:middle line:84% And as you will see, SNPs are the real meat 00:02:16.890 --> 00:02:20.020 align:middle line:90% of the human genome. 00:02:20.020 --> 00:02:25.350 align:middle line:84% So I will start from the '80s, when people started thinking 00:02:25.350 --> 00:02:26.700 align:middle line:90% about-- 00:02:26.700 --> 00:02:29.470 align:middle line:84% seriously thinking about developing a human genome 00:02:29.470 --> 00:02:29.970 align:middle line:90% project. 00:02:29.970 --> 00:02:32.430 align:middle line:90% And the Human Genome Project-- 00:02:32.430 --> 00:02:36.870 align:middle line:84% I will tell you what genetic polymorphisms are, 00:02:36.870 --> 00:02:39.570 align:middle line:84% what type of polymorphisms exist, and then 00:02:39.570 --> 00:02:42.930 align:middle line:84% we give you some basic terminology of what 00:02:42.930 --> 00:02:47.910 align:middle line:84% are the terms that geneticists use to describe 00:02:47.910 --> 00:02:50.040 align:middle line:90% the properties of a genome. 00:02:50.040 --> 00:02:52.740 align:middle line:84% And then I will talk to you about 00:02:52.740 --> 00:02:55.140 align:middle line:84% the real interesting stuff, which are complex traits, 00:02:55.140 --> 00:03:00.360 align:middle line:84% or those traits, those diseases, those observable features 00:03:00.360 --> 00:03:03.510 align:middle line:84% of an individual that don't come from one single gene, 00:03:03.510 --> 00:03:07.350 align:middle line:84% but they are the result of an interaction of more genes 00:03:07.350 --> 00:03:10.860 align:middle line:84% or the interaction of genes with the environment, 00:03:10.860 --> 00:03:14.405 align:middle line:84% and I will tell you why these traits are complex. 00:03:14.405 --> 00:03:16.590 align:middle line:84% And then I will tell you how people designed 00:03:16.590 --> 00:03:18.420 align:middle line:84% these kind of-- the experiments to identify 00:03:18.420 --> 00:03:21.750 align:middle line:84% these complex traits and also the simpler ones. 00:03:21.750 --> 00:03:23.280 align:middle line:84% And once they have designed it, I 00:03:23.280 --> 00:03:27.160 align:middle line:84% will tell you how do you analyze this data once you have them. 00:03:27.160 --> 00:03:29.820 align:middle line:84% And then I will tell you what's the word 00:03:29.820 --> 00:03:35.490 align:middle line:84% on the street on the latest fashion in analyzing 00:03:35.490 --> 00:03:38.300 align:middle line:90% the genome. 00:03:38.300 --> 00:03:44.350 align:middle line:84% So as I was saying, the intuition behind genetics 00:03:44.350 --> 00:03:48.280 align:middle line:84% is that we can find causes for diseases 00:03:48.280 --> 00:03:51.790 align:middle line:84% without knowing the actual mechanism of the disease, 00:03:51.790 --> 00:03:57.670 align:middle line:84% just associating changing in the genome to observable traits, 00:03:57.670 --> 00:03:59.590 align:middle line:90% observable characters. 00:03:59.590 --> 00:04:02.380 align:middle line:84% Now, as you can imagine, this is not really 00:04:02.380 --> 00:04:06.010 align:middle line:84% an easy idea to sell, and when people started selling it 00:04:06.010 --> 00:04:11.440 align:middle line:84% in the early 20th century, people didn't really 00:04:11.440 --> 00:04:14.570 align:middle line:84% like it a lot, especially biologists, who say, 00:04:14.570 --> 00:04:16.089 align:middle line:90% what kind of science is this? 00:04:16.089 --> 00:04:18.470 align:middle line:90% There is no mechanism in this. 00:04:18.470 --> 00:04:21.529 align:middle line:84% There is no understanding of how the things work. 00:04:21.529 --> 00:04:28.420 align:middle line:84% But then people started to deliver results, 00:04:28.420 --> 00:04:30.590 align:middle line:84% and how deliver results about this? 00:04:30.590 --> 00:04:34.210 align:middle line:84% Well, once we know that gene exists-- 00:04:34.210 --> 00:04:36.130 align:middle line:84% before, it was kind of difficult, but-- 00:04:36.130 --> 00:04:38.545 align:middle line:84% I will show you how you did it before, 00:04:38.545 --> 00:04:42.310 align:middle line:84% before knowing how to genotype somebody. 00:04:42.310 --> 00:04:45.220 align:middle line:84% But one of the-- the main intuition 00:04:45.220 --> 00:04:47.830 align:middle line:84% that changed everything is due to a guy called 00:04:47.830 --> 00:04:52.450 align:middle line:84% David Boston that, in 1997, showed that there are 00:04:52.450 --> 00:04:56.260 align:middle line:84% natural markers on the genome called polymorphisms 00:04:56.260 --> 00:05:00.700 align:middle line:84% that make us different and make this genome different, right? 00:05:00.700 --> 00:05:04.390 align:middle line:84% So if I have a little change in my population, what 00:05:04.390 --> 00:05:07.750 align:middle line:84% I can try to do is collect people that have the disease 00:05:07.750 --> 00:05:09.970 align:middle line:84% and collect people that don't have the disease 00:05:09.970 --> 00:05:12.760 align:middle line:84% and see which marker shows up more than I would 00:05:12.760 --> 00:05:16.525 align:middle line:90% expect in one group or another. 00:05:16.525 --> 00:05:18.400 align:middle line:84% And I will tell you how you play these games. 00:05:18.400 --> 00:05:20.650 align:middle line:84% You can play them in several ways, 00:05:20.650 --> 00:05:23.260 align:middle line:90% but the way in which we rely-- 00:05:23.260 --> 00:05:25.630 align:middle line:84% we do-- we conduct the studies is not 00:05:25.630 --> 00:05:29.680 align:middle line:84% by tagging something, but leveraging our natural tags 00:05:29.680 --> 00:05:31.420 align:middle line:84% where exactly these polymorphisms. 00:05:31.420 --> 00:05:34.040 align:middle line:90% 00:05:34.040 --> 00:05:35.870 align:middle line:84% This is just a summary of the central dogma 00:05:35.870 --> 00:05:37.280 align:middle line:90% of molecular biology. 00:05:37.280 --> 00:05:38.360 align:middle line:90% You all know this. 00:05:38.360 --> 00:05:42.380 align:middle line:84% You got a class from Dr. Butte, I think, about this. 00:05:42.380 --> 00:05:45.560 align:middle line:84% Some of you have a masters and PhDs in biology, 00:05:45.560 --> 00:05:48.650 align:middle line:84% so they know this better than I do. 00:05:48.650 --> 00:05:51.650 align:middle line:84% The assumption is that there is DNA 00:05:51.650 --> 00:05:53.720 align:middle line:84% that is all identical in all our cells, 00:05:53.720 --> 00:05:56.660 align:middle line:84% and then this is translated in RNA. 00:05:56.660 --> 00:06:00.170 align:middle line:84% And RNA is then turned into proteins 00:06:00.170 --> 00:06:04.175 align:middle line:84% which determines whatever we can observe of somebody, 00:06:04.175 --> 00:06:08.300 align:middle line:84% the traits, the diseases susceptible to, 00:06:08.300 --> 00:06:13.500 align:middle line:84% his physiology, the metabolism, even drug resistance. 00:06:13.500 --> 00:06:19.590 align:middle line:84% So the idea that polymorphisms could 00:06:19.590 --> 00:06:24.030 align:middle line:84% be used as natural markers created exactly the background 00:06:24.030 --> 00:06:26.650 align:middle line:84% of the Human Genome Project because, at that point, 00:06:26.650 --> 00:06:30.870 align:middle line:84% we had an intellectual tool, a scientific tool to track down 00:06:30.870 --> 00:06:33.150 align:middle line:90% diseases using the code. 00:06:33.150 --> 00:06:37.930 align:middle line:84% So the easiest thing we can do is 00:06:37.930 --> 00:06:41.020 align:middle line:84% to find out diseases that are caused 00:06:41.020 --> 00:06:45.382 align:middle line:84% by one single screw-up in one gene, 00:06:45.382 --> 00:06:47.590 align:middle line:84% and there are various ways in which this can actually 00:06:47.590 --> 00:06:50.230 align:middle line:90% be a screw-up or not. 00:06:50.230 --> 00:06:52.927 align:middle line:84% They can be dominant or recessive. 00:06:52.927 --> 00:06:54.760 align:middle line:84% Recessive means that you need to-- you know, 00:06:54.760 --> 00:06:57.850 align:middle line:84% you receive two chromosome, one from your father 00:06:57.850 --> 00:06:58.960 align:middle line:90% and one from your mother. 00:06:58.960 --> 00:07:00.550 align:middle line:84% To be recessive, you need to have 00:07:00.550 --> 00:07:02.590 align:middle line:84% both chromosomes that are screwed up, 00:07:02.590 --> 00:07:04.390 align:middle line:90% your gene in both chromosomes. 00:07:04.390 --> 00:07:06.980 align:middle line:84% To be dominant, one is going to be enough. 00:07:06.980 --> 00:07:11.320 align:middle line:84% And we classify these diseases also by another dimension, 00:07:11.320 --> 00:07:16.240 align:middle line:84% which is, if it's on the X chromosome, which is-- has-- 00:07:16.240 --> 00:07:19.090 align:middle line:90% there's a symmetry for men-- 00:07:19.090 --> 00:07:21.590 align:middle line:84% and/or if it's any other chromosome. 00:07:21.590 --> 00:07:24.640 align:middle line:84% So if it's any other chromosome, it's going to be an autosomal 00:07:24.640 --> 00:07:26.350 align:middle line:84% and can be dominant or recessive, 00:07:26.350 --> 00:07:28.120 align:middle line:84% and examples are the Huntington's disease 00:07:28.120 --> 00:07:29.350 align:middle line:90% and cystic fibrosis. 00:07:29.350 --> 00:07:34.000 align:middle line:84% And if it's X-linked, it can be dominant or recessive, 00:07:34.000 --> 00:07:37.120 align:middle line:84% and I will give you right away a little example of this. 00:07:37.120 --> 00:07:40.120 align:middle line:84% Today, there are about 400 single-gene diseases 00:07:40.120 --> 00:07:41.740 align:middle line:90% that have been identified. 00:07:41.740 --> 00:07:43.830 align:middle line:84% So for 400 diseases, we know that there 00:07:43.830 --> 00:07:45.010 align:middle line:90% is a little screw-up. 00:07:45.010 --> 00:07:48.010 align:middle line:84% Sometimes there is a very, very simple reason, 00:07:48.010 --> 00:07:54.280 align:middle line:84% because a base is turned from being 00:07:54.280 --> 00:07:58.340 align:middle line:84% a normal base, coding for amino acid, into stopping signals. 00:07:58.340 --> 00:08:01.360 align:middle line:84% So when that particular sequence is read, 00:08:01.360 --> 00:08:07.030 align:middle line:84% the protein does not code after that particular point 00:08:07.030 --> 00:08:10.645 align:middle line:84% and the protein is screwed up, and this creates disease. 00:08:10.645 --> 00:08:14.110 align:middle line:90% 00:08:14.110 --> 00:08:16.230 align:middle line:90% OK, so I need-- 00:08:16.230 --> 00:08:17.230 align:middle line:90% this is the boring part. 00:08:17.230 --> 00:08:19.630 align:middle line:90% I need to give you-- 00:08:19.630 --> 00:08:25.480 align:middle line:84% I need to give you a little terminology. 00:08:25.480 --> 00:08:29.950 align:middle line:84% Now, an allele is a sequence of DNA bases. 00:08:29.950 --> 00:08:33.590 align:middle line:90% 00:08:33.590 --> 00:08:40.049 align:middle line:84% So for each SNP, for each gene, for each whatever, 00:08:40.049 --> 00:08:43.110 align:middle line:84% for each piece of your DNA, you have two alleles, 00:08:43.110 --> 00:08:46.200 align:middle line:84% one coming from your mother and one from your father. 00:08:46.200 --> 00:08:50.050 align:middle line:84% Locus is a physical location on the chromosome. 00:08:50.050 --> 00:08:52.200 align:middle line:90% It's like [INAUDIBLE]. 00:08:52.200 --> 00:08:53.790 align:middle line:90% We know exactly where it is. 00:08:53.790 --> 00:08:55.320 align:middle line:90% We know it's there. 00:08:55.320 --> 00:08:57.420 align:middle line:84% We know on which chromosome is it, 00:08:57.420 --> 00:09:00.300 align:middle line:84% and we know the position of the chromosome. 00:09:00.300 --> 00:09:04.680 align:middle line:84% Linkage is a proximity of two alleles on the chromosome, 00:09:04.680 --> 00:09:08.430 align:middle line:84% and it is kind of ambiguous because, as you will see, 00:09:08.430 --> 00:09:12.570 align:middle line:84% proximity may have several meanings in genetics. 00:09:12.570 --> 00:09:16.620 align:middle line:84% The marker is an allele with a particular position 00:09:16.620 --> 00:09:19.440 align:middle line:90% that we can track down somehow. 00:09:19.440 --> 00:09:21.600 align:middle line:84% The distance, the physical distance, 00:09:21.600 --> 00:09:25.710 align:middle line:84% is the number of bases between one point and another, 00:09:25.710 --> 00:09:27.447 align:middle line:84% but this is not the only distance 00:09:27.447 --> 00:09:29.280 align:middle line:84% we have, and actually is not really the most 00:09:29.280 --> 00:09:32.100 align:middle line:90% interesting distance we have. 00:09:32.100 --> 00:09:34.590 align:middle line:84% The most interesting distance is the probability 00:09:34.590 --> 00:09:36.390 align:middle line:84% that two points on the chromosome 00:09:36.390 --> 00:09:41.160 align:middle line:84% will be recombined when you make your children. 00:09:41.160 --> 00:09:43.890 align:middle line:84% And the distance, this probabilistic distance, 00:09:43.890 --> 00:09:49.230 align:middle line:84% doesn't necessarily map on the chromosome 00:09:49.230 --> 00:09:52.740 align:middle line:84% with a constant number of bases, as you will see in a second. 00:09:52.740 --> 00:09:54.240 align:middle line:84% So there are-- this means that there 00:09:54.240 --> 00:09:56.820 align:middle line:84% are points on the chromosome that are easier to recombine 00:09:56.820 --> 00:10:00.120 align:middle line:84% and points on the chromosome that are harder to combine, 00:10:00.120 --> 00:10:02.880 align:middle line:90% less likely to recombine. 00:10:02.880 --> 00:10:08.010 align:middle line:84% The measure by which we identify this probabilistic distance 00:10:08.010 --> 00:10:09.960 align:middle line:90% is called centimorgan. 00:10:09.960 --> 00:10:12.780 align:middle line:84% A phenotype is a observable character. 00:10:12.780 --> 00:10:14.835 align:middle line:84% It can be susceptibility to a disease. 00:10:14.835 --> 00:10:18.930 align:middle line:84% It can be diabetes, a disease itself. 00:10:18.930 --> 00:10:22.320 align:middle line:84% The genotype is the internally coded inherited information, 00:10:22.320 --> 00:10:24.450 align:middle line:90% so it's a piece of DNA. 00:10:24.450 --> 00:10:30.870 align:middle line:84% And the penetrance is, if you take 00:10:30.870 --> 00:10:34.740 align:middle line:84% the frequent interpretation seriously, 00:10:34.740 --> 00:10:37.720 align:middle line:84% is the probability that, given that you have the allele, 00:10:37.720 --> 00:10:39.710 align:middle line:90% you will develop the phenotype. 00:10:39.710 --> 00:10:40.210 align:middle line:90% OK? 00:10:40.210 --> 00:10:44.710 align:middle line:90% 00:10:44.710 --> 00:10:47.640 align:middle line:84% So the physical distance between two alleles are a base pair, 00:10:47.640 --> 00:10:53.520 align:middle line:84% but the combination among them is not constant. 00:10:53.520 --> 00:10:55.590 align:middle line:84% Well, you all know Mendel's first law. 00:10:55.590 --> 00:10:57.870 align:middle line:84% We know that allele pair separate 00:10:57.870 --> 00:11:03.420 align:middle line:84% when the gamete is formed, and they are randomly reshuffled 00:11:03.420 --> 00:11:05.340 align:middle line:90% and create new pairs. 00:11:05.340 --> 00:11:09.300 align:middle line:84% Now, use-- the probabilistic measure we want to use 00:11:09.300 --> 00:11:12.150 align:middle line:84% is the probability that two points 00:11:12.150 --> 00:11:17.560 align:middle line:84% will be recombined at the next passage to the next generation. 00:11:17.560 --> 00:11:21.540 align:middle line:84% So we can say that, on average, 1 centimorgan, which 00:11:21.540 --> 00:11:26.610 align:middle line:84% is the probability of 1% of being recombined, is about-- 00:11:26.610 --> 00:11:28.560 align:middle line:90% happens between 1 megabases. 00:11:28.560 --> 00:11:30.570 align:middle line:84% Every 1 megabases, you have one probability 00:11:30.570 --> 00:11:33.780 align:middle line:90% of being recombined. 00:11:33.780 --> 00:11:37.350 align:middle line:84% Now, these are kind of little gossips. 00:11:37.350 --> 00:11:42.960 align:middle line:84% If human autosomal physical map is 3 billion bases, 00:11:42.960 --> 00:11:47.940 align:middle line:84% as you know, the linkage map in centimorgan-- the distance, 00:11:47.940 --> 00:11:49.890 align:middle line:84% probabilistic distance between points-- 00:11:49.890 --> 00:11:52.500 align:middle line:84% is different between humans and-- 00:11:52.500 --> 00:11:55.200 align:middle line:90% between male and female humans. 00:11:55.200 --> 00:11:58.320 align:middle line:84% There are 2,800 centimorgan for male 00:11:58.320 --> 00:12:02.590 align:middle line:84% and 4,200 centimorgan for females. 00:12:02.590 --> 00:12:03.090 align:middle line:90% Why's that? 00:12:03.090 --> 00:12:08.570 align:middle line:90% 00:12:08.570 --> 00:12:12.260 align:middle line:84% Females have a bigger genetic code. 00:12:12.260 --> 00:12:15.170 align:middle line:84% Male have this little Y wimpy chromosome, 00:12:15.170 --> 00:12:20.150 align:middle line:84% and females had this second big, beautiful X 00:12:20.150 --> 00:12:22.850 align:middle line:84% chromosome, which is much bigger than ours, 00:12:22.850 --> 00:12:26.810 align:middle line:90% less stable, but bigger. 00:12:26.810 --> 00:12:29.780 align:middle line:84% But if you compute the difference between these two 00:12:29.780 --> 00:12:32.990 align:middle line:84% things, you see that the probability of recombination 00:12:32.990 --> 00:12:37.070 align:middle line:84% of the correspondence between the probability 00:12:37.070 --> 00:12:39.710 align:middle line:84% of recombination for male and female is also different. 00:12:39.710 --> 00:12:45.260 align:middle line:84% One is a bigger code, and so the difference, you would say, 00:12:45.260 --> 00:12:46.340 align:middle line:90% is about 1-- 00:12:46.340 --> 00:12:49.010 align:middle line:90% on average, is 1 million pair-- 00:12:49.010 --> 00:12:51.170 align:middle line:84% 1 centimorgan for 1 million bases. 00:12:51.170 --> 00:12:55.220 align:middle line:84% And for female-- for male and female, it's about a bit more 00:12:55.220 --> 00:12:57.740 align:middle line:84% than a million for male and a bit less than a million 00:12:57.740 --> 00:12:59.600 align:middle line:90% for females. 00:12:59.600 --> 00:13:02.330 align:middle line:84% And one notion that would be absolutely 00:13:02.330 --> 00:13:05.330 align:middle line:84% important for our ability to use markers 00:13:05.330 --> 00:13:08.690 align:middle line:84% is of cosegregation, the fact that two alleles 00:13:08.690 --> 00:13:11.332 align:middle line:84% are transmitted together to the next generation. 00:13:11.332 --> 00:13:13.290 align:middle line:84% AUDIENCE: I just have a quick question on that. 00:13:13.290 --> 00:13:13.800 align:middle line:90% MARCO RAMONI: Sure. 00:13:13.800 --> 00:13:14.700 align:middle line:90% AUDIENCE: So why-- 00:13:14.700 --> 00:13:16.440 align:middle line:84% MARCO RAMONI: You're upset because they 00:13:16.440 --> 00:13:18.630 align:middle line:84% said that women had better chromosomes than we have? 00:13:18.630 --> 00:13:19.130 align:middle line:90% No. 00:13:19.130 --> 00:13:21.870 align:middle line:90% AUDIENCE: I agree. 00:13:21.870 --> 00:13:23.640 align:middle line:90% The question I had is-- 00:13:23.640 --> 00:13:27.540 align:middle line:84% was back in the definitions of a locus, and I 00:13:27.540 --> 00:13:29.670 align:middle line:90% just wondered, so the-- 00:13:29.670 --> 00:13:31.700 align:middle line:84% those things, can-- are those affected by, 00:13:31.700 --> 00:13:33.720 align:middle line:84% like-- so you said it's the physical location 00:13:33.720 --> 00:13:35.200 align:middle line:90% on a chromosome. 00:13:35.200 --> 00:13:37.000 align:middle line:84% Can the locus of a particular allele 00:13:37.000 --> 00:13:43.450 align:middle line:84% change if the packing into the chromosome structure 00:13:43.450 --> 00:13:44.508 align:middle line:90% is altered in any way? 00:13:44.508 --> 00:13:45.550 align:middle line:90% Like, is that something-- 00:13:45.550 --> 00:13:47.300 align:middle line:84% MARCO RAMONI: That's a very good question. 00:13:47.300 --> 00:13:47.838 align:middle line:90% So the-- 00:13:47.838 --> 00:13:50.380 align:middle line:84% AUDIENCE: And then follow-up to that is, if that is the case, 00:13:50.380 --> 00:13:52.858 align:middle line:84% does that affect the cosegregation as well? 00:13:52.858 --> 00:13:53.650 align:middle line:90% MARCO RAMONI: Yeah. 00:13:53.650 --> 00:13:56.150 align:middle line:90% So this is a very good question. 00:13:56.150 --> 00:13:59.350 align:middle line:84% You can imagine a locus as a variable and an allele 00:13:59.350 --> 00:14:01.250 align:middle line:90% as the state of this variable. 00:14:01.250 --> 00:14:03.640 align:middle line:84% So you have-- you are given this locus, 00:14:03.640 --> 00:14:10.480 align:middle line:84% and at position 3,022 on chromosome 5, you have an A. 00:14:10.480 --> 00:14:13.270 align:middle line:84% And if this is a polymorphism-- if it this is not 00:14:13.270 --> 00:14:20.080 align:middle line:84% a polymorphism, everybody has A. But if this is a polymorphism, 00:14:20.080 --> 00:14:25.510 align:middle line:84% we expect that something like above 1% of the population 00:14:25.510 --> 00:14:29.500 align:middle line:84% will have a different letter there, like a T or a G. 00:14:29.500 --> 00:14:35.050 align:middle line:84% Now, this is due to mutations, and there is nothing we can-- 00:14:35.050 --> 00:14:36.880 align:middle line:90% it's not affected by the locus. 00:14:36.880 --> 00:14:41.380 align:middle line:84% But the packaging of the genes, of the chromosomes, 00:14:41.380 --> 00:14:43.450 align:middle line:84% where they are next to each other, 00:14:43.450 --> 00:14:45.040 align:middle line:84% may affect the probability that it 00:14:45.040 --> 00:14:47.570 align:middle line:90% will be transmitted together. 00:14:47.570 --> 00:14:51.010 align:middle line:84% So there are several reasons why this linkage disequilibrium 00:14:51.010 --> 00:14:53.060 align:middle line:90% exists. 00:14:53.060 --> 00:14:54.850 align:middle line:84% One is the physical recombination 00:14:54.850 --> 00:14:58.480 align:middle line:84% of this, which is very difficult to identify, 00:14:58.480 --> 00:15:01.010 align:middle line:84% and the second are historical reasons. 00:15:01.010 --> 00:15:04.210 align:middle line:84% As I will tell you in a second, Caucasians 00:15:04.210 --> 00:15:07.900 align:middle line:84% come from a handful of people who left Africa, 00:15:07.900 --> 00:15:10.210 align:middle line:84% and this handful of people were basically 00:15:10.210 --> 00:15:12.290 align:middle line:90% assimilated in this process. 00:15:12.290 --> 00:15:15.640 align:middle line:84% So we all come from a very, very restricted number 00:15:15.640 --> 00:15:20.020 align:middle line:84% of people that were alive between 25,000 00:15:20.020 --> 00:15:22.330 align:middle line:84% and 50,000 years ago, which is a blink 00:15:22.330 --> 00:15:25.010 align:middle line:84% of an eye from an evolutionary point of view. 00:15:25.010 --> 00:15:32.560 align:middle line:84% So if you look at the map of an African American population, 00:15:32.560 --> 00:15:35.410 align:middle line:84% or an African population and a European population, 00:15:35.410 --> 00:15:37.780 align:middle line:84% you will see that there is a staggering difference 00:15:37.780 --> 00:15:40.570 align:middle line:84% in the variability of those and our variability. 00:15:40.570 --> 00:15:43.090 align:middle line:84% Now, if you look at our things, you 00:15:43.090 --> 00:15:45.760 align:middle line:84% may-- if you do a genotype study of Caucasians, 00:15:45.760 --> 00:15:48.380 align:middle line:84% you will see alleles that go together, 00:15:48.380 --> 00:15:52.030 align:middle line:84% but you will not be able to say if these alleles go together 00:15:52.030 --> 00:15:54.310 align:middle line:84% for actual physical recombination 00:15:54.310 --> 00:15:56.500 align:middle line:84% or simply because you live, we live-- 00:15:56.500 --> 00:16:01.210 align:middle line:84% we draw our genetic codes from a depleted pool in which only 00:16:01.210 --> 00:16:03.970 align:middle line:84% very few combinations are available. 00:16:03.970 --> 00:16:05.990 align:middle line:90% Does that answer your question? 00:16:05.990 --> 00:16:06.790 align:middle line:90% AUDIENCE: Yeah. 00:16:06.790 --> 00:16:08.260 align:middle line:84% MARCO RAMONI: I have a picture in a second that 00:16:08.260 --> 00:16:09.520 align:middle line:90% may be helpful in this. 00:16:09.520 --> 00:16:16.710 align:middle line:90% 00:16:16.710 --> 00:16:23.790 align:middle line:84% So let me go back to all this, a single disease works 00:16:23.790 --> 00:16:25.560 align:middle line:84% and how dramatic can be the effect 00:16:25.560 --> 00:16:27.360 align:middle line:90% that a single disease has. 00:16:27.360 --> 00:16:30.150 align:middle line:84% Hemophilia, as you know, is the X-linked recessive disease that 00:16:30.150 --> 00:16:32.070 align:middle line:90% is fatal for women, probably. 00:16:32.070 --> 00:16:34.530 align:middle line:90% So if a man-- 00:16:34.530 --> 00:16:36.750 align:middle line:90% it's on the X chromosome. 00:16:36.750 --> 00:16:40.150 align:middle line:84% We men don't get an X chromosome. 00:16:40.150 --> 00:16:44.140 align:middle line:90% So we only get recessive. 00:16:44.140 --> 00:16:46.530 align:middle line:90% We only get the mother. 00:16:46.530 --> 00:16:49.710 align:middle line:90% But-- and we-- 00:16:49.710 --> 00:16:53.895 align:middle line:84% so if the woman has it, she doesn't have any manifestation. 00:16:53.895 --> 00:16:58.620 align:middle line:84% She's just a carrier, and if a man has it, 00:16:58.620 --> 00:17:03.840 align:middle line:84% he has manifestations of the disease. 00:17:03.840 --> 00:17:07.470 align:middle line:90% We don't observe women with-- 00:17:07.470 --> 00:17:12.270 align:middle line:84% it's very rare to find women with both X 00:17:12.270 --> 00:17:13.980 align:middle line:90% chromosomes, male hemophilia. 00:17:13.980 --> 00:17:17.099 align:middle line:90% 00:17:17.099 --> 00:17:24.030 align:middle line:84% Now, this is a major screw-up in the history of Europe, 00:17:24.030 --> 00:17:27.510 align:middle line:90% caused by a gene like that. 00:17:27.510 --> 00:17:32.910 align:middle line:84% Now, this is Queen Victoria's family tree, 00:17:32.910 --> 00:17:36.150 align:middle line:90% and Queen Victoria is the one-- 00:17:36.150 --> 00:17:39.540 align:middle line:84% is the second-- is the one right in the middle, 00:17:39.540 --> 00:17:42.040 align:middle line:90% half orange and half white. 00:17:42.040 --> 00:17:45.720 align:middle line:84% So if they're round, both are women, and square are men. 00:17:45.720 --> 00:17:51.270 align:middle line:84% And if it's a half colored, she's a carrier. 00:17:51.270 --> 00:17:55.260 align:middle line:84% If it's all colored, is an affected child. 00:17:55.260 --> 00:18:00.720 align:middle line:84% So when Queen Victoria had her fifth child, 00:18:00.720 --> 00:18:02.790 align:middle line:90% but her first male-- 00:18:02.790 --> 00:18:04.170 align:middle line:90% we're talking about monarchy. 00:18:04.170 --> 00:18:06.990 align:middle line:84% Males have some kind of importance here. 00:18:06.990 --> 00:18:09.510 align:middle line:84% The first male, the male turned-- 00:18:09.510 --> 00:18:13.710 align:middle line:84% Leopold, the male, turned out to be hemophiliac. 00:18:13.710 --> 00:18:17.520 align:middle line:84% And she was very, very upset, and her declaration 00:18:17.520 --> 00:18:27.240 align:middle line:84% was that their blood was strong and there was no change in it, 00:18:27.240 --> 00:18:29.410 align:middle line:90% no weakness in it. 00:18:29.410 --> 00:18:32.880 align:middle line:84% So if we take her seriously, what 00:18:32.880 --> 00:18:34.860 align:middle line:90% would you think had happened? 00:18:34.860 --> 00:18:38.010 align:middle line:84% I mean, this thing has incredible consequences. 00:18:38.010 --> 00:18:43.698 align:middle line:84% If you look at down here and you see this row of seven-- 00:18:43.698 --> 00:18:47.430 align:middle line:84% of five children with no descendants, 00:18:47.430 --> 00:18:49.560 align:middle line:90% that's the Russian family. 00:18:49.560 --> 00:18:54.360 align:middle line:84% They were exterminated by the-- during the Russian Revolution, 00:18:54.360 --> 00:19:01.800 align:middle line:84% and people say that the wife of Nicholas II 00:19:01.800 --> 00:19:04.260 align:middle line:84% started seeing Rasputin-- was one 00:19:04.260 --> 00:19:07.230 align:middle line:84% of the major causes of the upsetting of the population-- 00:19:07.230 --> 00:19:13.170 align:middle line:84% because her first male born was hemophiliac, 00:19:13.170 --> 00:19:14.490 align:middle line:90% so a major screw-up. 00:19:14.490 --> 00:19:17.100 align:middle line:84% I mean, this led to the withdrawal of the Russians 00:19:17.100 --> 00:19:21.150 align:middle line:84% from the First World War and big, big, big changes. 00:19:21.150 --> 00:19:23.590 align:middle line:84% I mean we still pay taxes because they are communist, 00:19:23.590 --> 00:19:24.090 align:middle line:90% right? 00:19:24.090 --> 00:19:27.690 align:middle line:90% So what do you think happened? 00:19:27.690 --> 00:19:28.890 align:middle line:90% Who is responsible for it? 00:19:28.890 --> 00:19:31.500 align:middle line:90% 00:19:31.500 --> 00:19:33.800 align:middle line:84% They start from the assumption that they're strong, 00:19:33.800 --> 00:19:36.710 align:middle line:90% their blood is strong. 00:19:36.710 --> 00:19:41.630 align:middle line:84% You say, OK, so poor Victoria had a mutation. 00:19:41.630 --> 00:19:43.040 align:middle line:90% Would you believe that? 00:19:43.040 --> 00:19:45.770 align:middle line:90% Poor Leopold had the mutation? 00:19:45.770 --> 00:19:48.890 align:middle line:90% How would you compute that? 00:19:48.890 --> 00:19:51.290 align:middle line:84% What I would do is look at the randomness 00:19:51.290 --> 00:19:54.140 align:middle line:84% of the distribution of those alleles 00:19:54.140 --> 00:19:55.670 align:middle line:90% in the second generation. 00:19:55.670 --> 00:19:57.260 align:middle line:90% We cannot genotype these people. 00:19:57.260 --> 00:20:00.320 align:middle line:84% We can infer the genotype only by looking at what they have, 00:20:00.320 --> 00:20:01.490 align:middle line:90% right? 00:20:01.490 --> 00:20:03.830 align:middle line:90% So you would expect that-- 00:20:03.830 --> 00:20:09.920 align:middle line:84% all of these guys on the third row, 00:20:09.920 --> 00:20:13.400 align:middle line:90% they are Victoria's offspring. 00:20:13.400 --> 00:20:17.360 align:middle line:84% You would expect that 50/50 distribution, right? 00:20:17.360 --> 00:20:19.550 align:middle line:84% Now, for some of them, we cannot really evaluate. 00:20:19.550 --> 00:20:21.350 align:middle line:84% We have two-- we have three we have to remove out 00:20:21.350 --> 00:20:22.400 align:middle line:84% because they don't have the same, 00:20:22.400 --> 00:20:24.090 align:middle line:84% and we don't know anything about that. 00:20:24.090 --> 00:20:27.830 align:middle line:84% If you take the other, we have 1, 2-- 00:20:27.830 --> 00:20:29.780 align:middle line:90% we have 1, 2-- 00:20:29.780 --> 00:20:37.550 align:middle line:84% they are actually healthy, and we can almost say that. 00:20:37.550 --> 00:20:42.650 align:middle line:84% And then we have three that are either affected or carriers. 00:20:42.650 --> 00:20:45.560 align:middle line:90% This is a good 50%, right? 00:20:45.560 --> 00:20:50.520 align:middle line:84% I mean, I would say that you cannot really get any better 00:20:50.520 --> 00:20:52.190 align:middle line:90% than that with this sample size. 00:20:52.190 --> 00:20:54.350 align:middle line:90% It's almost a perfect 50%. 00:20:54.350 --> 00:20:56.150 align:middle line:84% If you go down and look down streams, 00:20:56.150 --> 00:20:58.430 align:middle line:84% you will see that the ratio will be exactly the same. 00:20:58.430 --> 00:21:02.390 align:middle line:84% For the women-- for the daughters of Nicholas II, 00:21:02.390 --> 00:21:07.410 align:middle line:84% we cannot know, but if you look at R in there, you will see, 00:21:07.410 --> 00:21:09.450 align:middle line:90% you have the same distribution. 00:21:09.450 --> 00:21:12.230 align:middle line:84% So I would say that from Victoria onward, 00:21:12.230 --> 00:21:14.940 align:middle line:90% everything is pretty consistent. 00:21:14.940 --> 00:21:19.430 align:middle line:84% So she didn't-- so, OK, she's might have a mutation, 00:21:19.430 --> 00:21:26.383 align:middle line:84% or she may have been cheating on her husband? 00:21:26.383 --> 00:21:27.800 align:middle line:84% No, because otherwise she would've 00:21:27.800 --> 00:21:30.350 align:middle line:84% been cheating with everybody, right, for all the children. 00:21:30.350 --> 00:21:33.770 align:middle line:84% We have a very pretty nice distribution of this. 00:21:33.770 --> 00:21:35.115 align:middle line:90% So what happened? 00:21:35.115 --> 00:21:35.990 align:middle line:90% How about her mother? 00:21:35.990 --> 00:21:40.163 align:middle line:90% 00:21:40.163 --> 00:21:42.580 align:middle line:84% Well, her mother, we don't have really enough information. 00:21:42.580 --> 00:21:44.500 align:middle line:90% How about her father? 00:21:44.500 --> 00:21:49.250 align:middle line:84% Her father married a woman 20 years younger than he was. 00:21:49.250 --> 00:21:54.640 align:middle line:84% He was rumored to be homosexual, and he died six months later 00:21:54.640 --> 00:21:55.600 align:middle line:90% Victoria was born. 00:21:55.600 --> 00:21:58.530 align:middle line:90% 00:21:58.530 --> 00:22:01.500 align:middle line:90% The postman maybe? 00:22:01.500 --> 00:22:03.300 align:middle line:84% If you have an hemophiliac postman, 00:22:03.300 --> 00:22:05.880 align:middle line:84% there would be a good explanation. 00:22:05.880 --> 00:22:09.170 align:middle line:84% Now, somebody made some research and found out 00:22:09.170 --> 00:22:16.200 align:middle line:84% that a grandmother of Queen Victoria's had 00:22:16.200 --> 00:22:20.040 align:middle line:84% two affected siblings who died of hemophilia. 00:22:20.040 --> 00:22:23.460 align:middle line:84% So in this case, the blood is not really that strong 00:22:23.460 --> 00:22:29.070 align:middle line:84% because it's coming in from Saxe-Coburg, from that line, 00:22:29.070 --> 00:22:36.282 align:middle line:84% and becomes part of the monarchy of the European aristocracy. 00:22:36.282 --> 00:22:37.740 align:middle line:84% You don't like this kind of gossip? 00:22:37.740 --> 00:22:38.460 align:middle line:90% I mean, this is-- 00:22:38.460 --> 00:22:39.825 align:middle line:90% [LAUGHTER] 00:22:39.825 --> 00:22:41.190 align:middle line:90% 00:22:41.190 --> 00:22:43.530 align:middle line:84% I mean, this was really the story of the day, 00:22:43.530 --> 00:22:45.270 align:middle line:90% of the century, in-- 00:22:45.270 --> 00:22:46.680 align:middle line:90% at that time. 00:22:46.680 --> 00:22:48.310 align:middle line:90% But you see how we did it. 00:22:48.310 --> 00:22:49.180 align:middle line:90% We had to wait 200-- 00:22:49.180 --> 00:22:51.840 align:middle line:84% a hundred and something years to find out 00:22:51.840 --> 00:22:54.400 align:middle line:84% what really happened because we had to observe-- 00:22:54.400 --> 00:22:56.575 align:middle line:84% we cannot genotype these people at that time, 00:22:56.575 --> 00:22:58.200 align:middle line:84% so we have to observe their, of course, 00:22:58.200 --> 00:23:00.490 align:middle line:90% characters of their things. 00:23:00.490 --> 00:23:04.030 align:middle line:84% But if we were able to genotype them, what could we do? 00:23:04.030 --> 00:23:07.200 align:middle line:84% Well, we could look at this thing of these small variations 00:23:07.200 --> 00:23:10.920 align:middle line:84% in their chromosomes, and the oldest variation 00:23:10.920 --> 00:23:15.360 align:middle line:84% we have been able to use are called simple sequence repeats 00:23:15.360 --> 00:23:16.830 align:middle line:90% and microsatellites. 00:23:16.830 --> 00:23:20.940 align:middle line:84% So these are-- let me see if I have a slide explaining this. 00:23:20.940 --> 00:23:21.810 align:middle line:90% No. 00:23:21.810 --> 00:23:25.770 align:middle line:84% So there are parts of the chromosomes of the genetic 00:23:25.770 --> 00:23:30.360 align:middle line:84% code, in which you have sequences like G-A-T-A that are 00:23:30.360 --> 00:23:34.530 align:middle line:84% repeated several times in your chromosome, 00:23:34.530 --> 00:23:35.790 align:middle line:90% in your genetic code. 00:23:35.790 --> 00:23:39.810 align:middle line:84% They are repeated 13 time in me and 15 times in you, 00:23:39.810 --> 00:23:42.430 align:middle line:90% two times in somebody else. 00:23:42.430 --> 00:23:45.900 align:middle line:84% And if we count these things, we can actually 00:23:45.900 --> 00:23:49.890 align:middle line:84% identify a region of the chromosome that by linkage-- 00:23:49.890 --> 00:23:55.200 align:middle line:84% remember, by being attached to it from evolutionary purposes-- 00:23:55.200 --> 00:23:57.930 align:middle line:84% we'll identify a part, a stretch of the genome 00:23:57.930 --> 00:23:59.370 align:middle line:90% that we can actually tag. 00:23:59.370 --> 00:24:03.510 align:middle line:84% Now, the problem is that simple sequence repeats 00:24:03.510 --> 00:24:09.870 align:middle line:84% and microsatellites add this little drawback. 00:24:09.870 --> 00:24:13.650 align:middle line:84% They cannot occur in really interesting regions. 00:24:13.650 --> 00:24:19.080 align:middle line:84% You cannot change to match a genetic code by repeating 00:24:19.080 --> 00:24:23.250 align:middle line:84% the same thing 13 times or two times and hope to get coherent 00:24:23.250 --> 00:24:25.500 align:middle line:90% proteins out of it, right? 00:24:25.500 --> 00:24:28.770 align:middle line:84% So we are confined in interest in regions 00:24:28.770 --> 00:24:30.660 align:middle line:84% that, by design, are supposed to be 00:24:30.660 --> 00:24:32.817 align:middle line:90% kind of uninteresting from an-- 00:24:32.817 --> 00:24:34.150 align:middle line:90% from a functional point of view. 00:24:34.150 --> 00:24:36.570 align:middle line:84% There may be interest from an evolutionary point of view, 00:24:36.570 --> 00:24:40.320 align:middle line:84% but they're kind of leftover of something else. 00:24:40.320 --> 00:24:43.020 align:middle line:84% Now, SNPs have this great property. 00:24:43.020 --> 00:24:45.120 align:middle line:90% They are one single base. 00:24:45.120 --> 00:24:49.020 align:middle line:90% And so it's something like this. 00:24:49.020 --> 00:24:52.920 align:middle line:84% You have your sequence, and you have that little base that 00:24:52.920 --> 00:24:57.960 align:middle line:84% in 90% of the population is T and in 10% of the population 00:24:57.960 --> 00:24:59.370 align:middle line:90% is G. 00:24:59.370 --> 00:25:01.650 align:middle line:90% To be sure that-- 00:25:01.650 --> 00:25:05.550 align:middle line:84% to qualify a single change like this as a SNP 00:25:05.550 --> 00:25:07.500 align:middle line:84% and not just a simple mutation, it 00:25:07.500 --> 00:25:10.350 align:middle line:84% has to have a minimal appearance in the population. 00:25:10.350 --> 00:25:13.650 align:middle line:84% And we stipulated this 1%, but nobody takes the number 00:25:13.650 --> 00:25:16.830 align:middle line:84% seriously because that's below the genotype error-- 00:25:16.830 --> 00:25:17.920 align:middle line:90% the expected error. 00:25:17.920 --> 00:25:20.520 align:middle line:84% So when you genotype people, usually you 00:25:20.520 --> 00:25:23.760 align:middle line:84% have a higher cutoff to say, well, this is not really a SNP. 00:25:23.760 --> 00:25:26.790 align:middle line:84% This is something that is just due to genotype error. 00:25:26.790 --> 00:25:29.370 align:middle line:90% 00:25:29.370 --> 00:25:32.220 align:middle line:84% They are also the most common type of variations. 00:25:32.220 --> 00:25:34.570 align:middle line:84% Not only they go in interesting places, 00:25:34.570 --> 00:25:38.160 align:middle line:84% but they can actually be in any place in the genome. 00:25:38.160 --> 00:25:43.560 align:middle line:84% And they have different functions and different roles 00:25:43.560 --> 00:25:47.560 align:middle line:90% in terms of protein coding. 00:25:47.560 --> 00:25:50.700 align:middle line:84% So a cSNP is a SNP occurring in a coding region. 00:25:50.700 --> 00:25:55.180 align:middle line:84% An rSNP is a SNP occurring in a regulatory region. 00:25:55.180 --> 00:25:59.080 align:middle line:84% The sSNP is the gene that occurs in a coding region, 00:25:59.080 --> 00:26:03.300 align:middle line:84% but by changing the SNP, you don't change the amino acid. 00:26:03.300 --> 00:26:05.940 align:middle line:84% Right, we have all this redundant vocabulary 00:26:05.940 --> 00:26:08.890 align:middle line:84% of amino acids, and in this case, 00:26:08.890 --> 00:26:11.560 align:middle line:84% it's going to be functionally silent, 00:26:11.560 --> 00:26:21.790 align:middle line:84% but it's going still to be a marker. 00:26:21.790 --> 00:26:25.125 align:middle line:84% Now, why actually we cannot do this-- 00:26:25.125 --> 00:26:27.810 align:middle line:90% 00:26:27.810 --> 00:26:32.100 align:middle line:84% why SNPs are important, and why we cannot have microsatellites 00:26:32.100 --> 00:26:33.180 align:middle line:90% in other regions? 00:26:33.180 --> 00:26:36.900 align:middle line:84% Well, this is a very old study, one of the very first. 00:26:36.900 --> 00:26:39.210 align:middle line:84% And what this Kreitman guy did was 00:26:39.210 --> 00:26:43.380 align:middle line:84% to sequence 11 alleles from a locus 00:26:43.380 --> 00:26:48.930 align:middle line:84% from a gene called alcohol dehydrogenase into drosophila. 00:26:48.930 --> 00:26:55.020 align:middle line:84% Now, if you-- you have 11 coding regions, 00:26:55.020 --> 00:26:58.730 align:middle line:84% and you have 14 sites that have alternative bases. 00:26:58.730 --> 00:27:03.610 align:middle line:84% Now, if you simply imagine that these are random changes, 00:27:03.610 --> 00:27:08.920 align:middle line:84% you would expect that about 70%, 75% of them 00:27:08.920 --> 00:27:12.680 align:middle line:90% would change the amino acid. 00:27:12.680 --> 00:27:14.680 align:middle line:84% Now, when you actually look at them, 00:27:14.680 --> 00:27:19.180 align:middle line:84% you see that basically none of them does. 00:27:19.180 --> 00:27:21.170 align:middle line:90% Why is that? 00:27:21.170 --> 00:27:25.600 align:middle line:84% Well, because this is a very important gene for this animal. 00:27:25.600 --> 00:27:33.190 align:middle line:84% They are born and they are nurtured in alcohol, like some 00:27:33.190 --> 00:27:34.750 align:middle line:90% of my friends in college. 00:27:34.750 --> 00:27:39.730 align:middle line:84% And they-- so the ability to detoxify alcohol 00:27:39.730 --> 00:27:45.350 align:middle line:84% is going to be a very important evolutionary point for them. 00:27:45.350 --> 00:27:49.570 align:middle line:84% So nature can make mistake, but it 00:27:49.570 --> 00:27:54.910 align:middle line:84% can give them so many changes in a critical region. 00:27:54.910 --> 00:28:01.570 align:middle line:84% But I want to challenge a little your Darwinian souls. 00:28:01.570 --> 00:28:05.530 align:middle line:84% Mind, it's not that we observe them, 00:28:05.530 --> 00:28:08.080 align:middle line:84% we have the change, the random change, 00:28:08.080 --> 00:28:12.520 align:middle line:84% we kill off 75% of them, and then 00:28:12.520 --> 00:28:16.300 align:middle line:84% we see that the remaining 25% have survived. 00:28:16.300 --> 00:28:18.880 align:middle line:90% These things just don't happen. 00:28:18.880 --> 00:28:23.110 align:middle line:84% There is no natural selection, in terms of environment, 00:28:23.110 --> 00:28:25.120 align:middle line:90% that is killing the 75 people-- 00:28:25.120 --> 00:28:28.130 align:middle line:90% 75% of drosophila, OK? 00:28:28.130 --> 00:28:31.770 align:middle line:90% 00:28:31.770 --> 00:28:33.930 align:middle line:90% You with me? 00:28:33.930 --> 00:28:36.630 align:middle line:84% So in a Darwinian scheme, what we would expect 00:28:36.630 --> 00:28:38.490 align:middle line:84% is that you have a random mutation. 00:28:38.490 --> 00:28:39.460 align:middle line:90% You go out. 00:28:39.460 --> 00:28:40.710 align:middle line:90% You don't run fast enough. 00:28:40.710 --> 00:28:42.930 align:middle line:90% You get killed. 00:28:42.930 --> 00:28:45.900 align:middle line:84% In this case, you get a random mutation. 00:28:45.900 --> 00:28:50.160 align:middle line:84% Your random mutation is selected by the environment in some way, 00:28:50.160 --> 00:28:52.950 align:middle line:84% typically by killing you, and then-- 00:28:52.950 --> 00:28:56.020 align:middle line:84% before you can reproduce-- and that's it. 00:28:56.020 --> 00:29:01.150 align:middle line:84% But in this case, these things just don't happen, 00:29:01.150 --> 00:29:05.370 align:middle line:84% so there is no random change and then selection. 00:29:05.370 --> 00:29:09.810 align:middle line:84% There is something there in the control mechanism 00:29:09.810 --> 00:29:12.750 align:middle line:84% that will prevent the animal to have the mutation 00:29:12.750 --> 00:29:15.350 align:middle line:90% to begin with. 00:29:15.350 --> 00:29:15.850 align:middle line:90% OK? 00:29:15.850 --> 00:29:20.520 align:middle line:90% 00:29:20.520 --> 00:29:23.170 align:middle line:84% This is the SNP map I was saying before. 00:29:23.170 --> 00:29:25.570 align:middle line:90% This is how you read a SNP map. 00:29:25.570 --> 00:29:30.150 align:middle line:84% So you remember that you have two alleles for each locus, 00:29:30.150 --> 00:29:32.160 align:middle line:84% and in this case, what we have done 00:29:32.160 --> 00:29:35.580 align:middle line:84% is to genotype all these people-- 00:29:35.580 --> 00:29:37.080 align:middle line:90% they are the rows-- 00:29:37.080 --> 00:29:40.570 align:middle line:84% for these loci, that are the columns. 00:29:40.570 --> 00:29:43.170 align:middle line:84% So if an individual has a blue spot, 00:29:43.170 --> 00:29:47.410 align:middle line:84% the blue spot means that he is homozygous at the major allele, 00:29:47.410 --> 00:29:52.680 align:middle line:84% so he has both chromosomes with the most frequent allele, 00:29:52.680 --> 00:29:54.180 align:middle line:84% the most frequent in the population. 00:29:54.180 --> 00:29:57.330 align:middle line:84% If he's green, he's heterozygous, and if he's red, 00:29:57.330 --> 00:29:59.730 align:middle line:84% he's homozygous in the minor alleles. 00:29:59.730 --> 00:30:02.220 align:middle line:90% And his wife is a missing link. 00:30:02.220 --> 00:30:04.590 align:middle line:84% Now, how would you do an association study? 00:30:04.590 --> 00:30:06.900 align:middle line:84% Suppose I draw that red line and I tell you, 00:30:06.900 --> 00:30:09.540 align:middle line:84% you know, down here, I have the phenotype. 00:30:09.540 --> 00:30:11.550 align:middle line:84% Up there, I don't have the phenotype. 00:30:11.550 --> 00:30:13.200 align:middle line:90% Could you make some association? 00:30:13.200 --> 00:30:14.980 align:middle line:84% Well, you know, if you look here, 00:30:14.980 --> 00:30:17.340 align:middle line:84% there are some genes that, one side are all blue 00:30:17.340 --> 00:30:19.350 align:middle line:90% and one side are more colorful. 00:30:19.350 --> 00:30:22.210 align:middle line:84% You can do some statistics and say, well, 00:30:22.210 --> 00:30:25.890 align:middle line:84% this thing would allow me to distinguish between these two 00:30:25.890 --> 00:30:30.690 align:middle line:84% groups and predict which one is going to get the SNP-- 00:30:30.690 --> 00:30:31.750 align:middle line:90% the phenotype. 00:30:31.750 --> 00:30:35.010 align:middle line:84% Now, I'll ask you a slightly different question. 00:30:35.010 --> 00:30:36.720 align:middle line:90% This is your SNP map. 00:30:36.720 --> 00:30:38.010 align:middle line:90% What is the phenotype? 00:30:38.010 --> 00:30:43.400 align:middle line:90% 00:30:43.400 --> 00:30:48.140 align:middle line:84% What makes these two groups different, something 00:30:48.140 --> 00:30:50.860 align:middle line:90% that is observable? 00:30:50.860 --> 00:30:51.580 align:middle line:90% Can you say that? 00:30:51.580 --> 00:30:59.870 align:middle line:90% 00:30:59.870 --> 00:31:01.310 align:middle line:90% Come on. 00:31:01.310 --> 00:31:03.500 align:middle line:90% Buy a pizza. 00:31:03.500 --> 00:31:04.070 align:middle line:90% Guess. 00:31:04.070 --> 00:31:04.310 align:middle line:90% Why? 00:31:04.310 --> 00:31:04.810 align:middle line:90% Guess. 00:31:04.810 --> 00:31:11.598 align:middle line:90% 00:31:11.598 --> 00:31:14.200 align:middle line:90% No taker? 00:31:14.200 --> 00:31:15.850 align:middle line:84% AUDIENCE: This is a human map or-- 00:31:15.850 --> 00:31:17.392 align:middle line:84% MARCO RAMONI: Yeah, it's a human map. 00:31:17.392 --> 00:31:19.270 align:middle line:90% These are two human samples. 00:31:19.270 --> 00:31:21.817 align:middle line:84% AUDIENCE: I mean, [INAUDIBLE] African Americans. 00:31:21.817 --> 00:31:23.150 align:middle line:90% MARCO RAMONI: You got the pizza. 00:31:23.150 --> 00:31:25.070 align:middle line:90% [LAUGHTER] 00:31:25.070 --> 00:31:27.110 align:middle line:84% Down there are African Americans. 00:31:27.110 --> 00:31:30.920 align:middle line:90% Up there are Caucasians. 00:31:30.920 --> 00:31:37.640 align:middle line:84% And so to go back to your question, if you look-- 00:31:37.640 --> 00:31:43.130 align:middle line:84% so the evolutionary pressure of the link between two elements 00:31:43.130 --> 00:31:48.110 align:middle line:84% in the European population-- take the fifth and the-- 00:31:48.110 --> 00:31:52.160 align:middle line:84% no, take the sixth and the seventh, two spots, right? 00:31:52.160 --> 00:31:59.940 align:middle line:84% There you-- no, let me do this this way. 00:31:59.940 --> 00:32:03.180 align:middle line:90% Take these two points, OK? 00:32:03.180 --> 00:32:04.680 align:middle line:84% These two points, if you study them, 00:32:04.680 --> 00:32:07.020 align:middle line:84% they go together, if you look at the statistics, 00:32:07.020 --> 00:32:08.800 align:middle line:90% they will go together. 00:32:08.800 --> 00:32:12.010 align:middle line:84% Why-- if you do it in a European population. 00:32:12.010 --> 00:32:12.510 align:middle line:90% Why? 00:32:12.510 --> 00:32:15.580 align:middle line:84% Because you don't have alternatives. 00:32:15.580 --> 00:32:20.160 align:middle line:84% You don't know if these things really go together because some 00:32:20.160 --> 00:32:23.380 align:middle line:84% physical reason, or they're really just recombining like 00:32:23.380 --> 00:32:26.970 align:middle line:84% crazy but we cannot observe because our population is 00:32:26.970 --> 00:32:30.390 align:middle line:84% a population that doesn't have that particular SNP and this 00:32:30.390 --> 00:32:33.010 align:middle line:84% SNP exists only in this other population. 00:32:33.010 --> 00:32:36.610 align:middle line:84% In this case, it's the African American people. 00:32:36.610 --> 00:32:39.067 align:middle line:90% Does this answer your question? 00:32:39.067 --> 00:32:40.650 align:middle line:84% AUDIENCE: I'll have to think about it. 00:32:40.650 --> 00:32:41.358 align:middle line:90% MARCO RAMONI: OK. 00:32:41.358 --> 00:32:46.060 align:middle line:90% 00:32:46.060 --> 00:32:48.340 align:middle line:90% So let me go back here. 00:32:48.340 --> 00:32:52.180 align:middle line:90% 00:32:52.180 --> 00:32:57.950 align:middle line:84% So the first quantitative law we-- the most important 00:32:57.950 --> 00:33:01.280 align:middle line:84% quantitative law we have in genetics 00:33:01.280 --> 00:33:08.720 align:middle line:84% dictates how many major, minor, and heterozygous people we'll 00:33:08.720 --> 00:33:11.730 align:middle line:90% have in a particular population. 00:33:11.730 --> 00:33:13.940 align:middle line:90% Let me-- come on. 00:33:13.940 --> 00:33:14.690 align:middle line:90% OK. 00:33:14.690 --> 00:33:17.450 align:middle line:84% So in this case, you take a single allele, 00:33:17.450 --> 00:33:19.820 align:middle line:84% and you can actually see that you 00:33:19.820 --> 00:33:23.450 align:middle line:84% have a-- you expect to have more people with the homozygous 00:33:23.450 --> 00:33:27.140 align:middle line:84% at the major allele, less people heterozygous, and less 00:33:27.140 --> 00:33:32.660 align:middle line:84% people that are homozygous at the minor allele. 00:33:32.660 --> 00:33:37.190 align:middle line:84% Now, what rule exists there to make-- to distribute 00:33:37.190 --> 00:33:39.230 align:middle line:90% these proportions? 00:33:39.230 --> 00:33:41.830 align:middle line:84% So the law is the Hardy-Weinberg law. 00:33:41.830 --> 00:33:43.910 align:middle line:84% It states that the probability of having 00:33:43.910 --> 00:33:49.130 align:middle line:84% a major allele, the minor allele, and an heterozygous is 00:33:49.130 --> 00:33:56.540 align:middle line:84% this formula, is p squared plus 2pq plus p squared equals 1. 00:33:56.540 --> 00:34:00.140 align:middle line:84% We call this a situation of equilibrium. 00:34:00.140 --> 00:34:02.690 align:middle line:84% When everything is all right in the population, 00:34:02.690 --> 00:34:05.690 align:middle line:84% there hasn't been some major screw-up, 00:34:05.690 --> 00:34:08.070 align:middle line:84% this is the law that dictates the distribution. 00:34:08.070 --> 00:34:11.130 align:middle line:84% This is what we expect in a population in equilibrium. 00:34:11.130 --> 00:34:12.980 align:middle line:84% Now, in a hermaphroditic population 00:34:12.980 --> 00:34:14.810 align:middle line:84% we have an equilibrium in one generation 00:34:14.810 --> 00:34:17.449 align:middle line:90% by redistributing these things. 00:34:17.449 --> 00:34:22.020 align:middle line:84% We have no autosomal genes, so in our case, 00:34:22.020 --> 00:34:25.889 align:middle line:84% we need two generations to get an equilibrium. 00:34:25.889 --> 00:34:29.449 align:middle line:84% Now, you can use this to make a lot of little games, 00:34:29.449 --> 00:34:32.510 align:middle line:84% like, for instance, how many Caucasians are 00:34:32.510 --> 00:34:35.960 align:middle line:90% carriers of cystic fibrosis? 00:34:35.960 --> 00:34:37.639 align:middle line:84% Well, we know that cystic fibrosis 00:34:37.639 --> 00:34:45.230 align:middle line:84% affects 1 of 2,500 Caucasians, so this is our q square, right? 00:34:45.230 --> 00:34:50.540 align:middle line:84% So 0.02 is going to be our q, and the number 00:34:50.540 --> 00:34:53.090 align:middle line:84% of nonaffected alleles is going to be 98-- 00:34:53.090 --> 00:34:56.090 align:middle line:90% 0.98, 98%, right? 00:34:56.090 --> 00:35:02.900 align:middle line:84% So we can plug p and q up there and compute the 2pq as 1 00:35:02.900 --> 00:35:07.940 align:middle line:84% out of 25, which is a remarkably high number 00:35:07.940 --> 00:35:09.170 align:middle line:90% for a disease like this. 00:35:09.170 --> 00:35:11.820 align:middle line:90% 00:35:11.820 --> 00:35:13.920 align:middle line:84% Now, we make all these calculations 00:35:13.920 --> 00:35:16.810 align:middle line:90% under some assumptions. 00:35:16.810 --> 00:35:18.790 align:middle line:84% First assumption we make is random mating, 00:35:18.790 --> 00:35:21.690 align:middle line:84% which is not a justification to have sex with anybody 00:35:21.690 --> 00:35:24.930 align:middle line:84% but is the mechanism of reproduction 00:35:24.930 --> 00:35:28.140 align:middle line:84% that we imagine that exists in our population. 00:35:28.140 --> 00:35:30.870 align:middle line:84% That is, the person you will mate with 00:35:30.870 --> 00:35:36.010 align:middle line:84% will not select you on the basis of a particular genotype. 00:35:36.010 --> 00:35:39.330 align:middle line:84% If we are interested in that particular genotype, 00:35:39.330 --> 00:35:43.050 align:middle line:84% she would not ask you for your genetic map 00:35:43.050 --> 00:35:46.314 align:middle line:84% and say, no, you have a T in this locus, I don't want you. 00:35:46.314 --> 00:35:47.970 align:middle line:90% OK? 00:35:47.970 --> 00:35:51.690 align:middle line:84% If this is the case, all our calculations go down the drain. 00:35:51.690 --> 00:35:53.520 align:middle line:84% Well, we can make some adjustment then. 00:35:53.520 --> 00:35:57.310 align:middle line:90% 00:35:57.310 --> 00:35:59.230 align:middle line:84% The consequence of this is that we 00:35:59.230 --> 00:36:01.930 align:middle line:84% have a lot of problems when the selection is 00:36:01.930 --> 00:36:05.200 align:middle line:84% based on things that are proxies, 00:36:05.200 --> 00:36:08.080 align:middle line:90% like being my relative. 00:36:08.080 --> 00:36:10.480 align:middle line:84% If somebody is my relative, I'm going 00:36:10.480 --> 00:36:13.570 align:middle line:84% to introduce a bias because I'm going 00:36:13.570 --> 00:36:17.710 align:middle line:84% to select her on the basis of her genetic code. 00:36:17.710 --> 00:36:19.960 align:middle line:84% I know I don't do this for the genetic code, 00:36:19.960 --> 00:36:21.070 align:middle line:90% but I'm going to have-- 00:36:21.070 --> 00:36:23.830 align:middle line:90% I'm going to do it anyway. 00:36:23.830 --> 00:36:28.450 align:middle line:84% And the other assumption we have is 00:36:28.450 --> 00:36:30.310 align:middle line:84% that we have an infinite population, which 00:36:30.310 --> 00:36:32.560 align:middle line:84% is kind of sensible for us, for 6 billion people. 00:36:32.560 --> 00:36:38.570 align:middle line:90% 00:36:38.570 --> 00:36:40.420 align:middle line:90% So the other thing we-- 00:36:40.420 --> 00:36:42.210 align:middle line:84% so why do we need these assumptions? 00:36:42.210 --> 00:36:49.330 align:middle line:84% Well, because we know that the mechanism of spread 00:36:49.330 --> 00:36:51.430 align:middle line:84% of a character in the population-- 00:36:51.430 --> 00:36:52.730 align:middle line:90% it's called drift-- 00:36:52.730 --> 00:36:56.440 align:middle line:84% is the way in which a particular allele 00:36:56.440 --> 00:37:00.940 align:middle line:84% will get into the population and either conquer it all or just 00:37:00.940 --> 00:37:03.730 align:middle line:90% simply disappear, right? 00:37:03.730 --> 00:37:05.590 align:middle line:84% So at some point we will reach some kind 00:37:05.590 --> 00:37:07.300 align:middle line:84% of evolutionary stability in which 00:37:07.300 --> 00:37:08.890 align:middle line:90% these things will be around. 00:37:08.890 --> 00:37:13.600 align:middle line:84% And the problem with this, the moment in which people in the-- 00:37:13.600 --> 00:37:15.430 align:middle line:84% mate locally, because, again, this 00:37:15.430 --> 00:37:17.320 align:middle line:84% creates some particular population-- 00:37:17.320 --> 00:37:19.900 align:middle line:84% some particular bias with the fact 00:37:19.900 --> 00:37:23.440 align:middle line:84% that your fishing pond is too small. 00:37:23.440 --> 00:37:27.280 align:middle line:84% So at the end, you are going to erase everything 00:37:27.280 --> 00:37:29.080 align:middle line:90% in that particular population. 00:37:29.080 --> 00:37:31.750 align:middle line:84% If everybody mates with each other in the particular-- 00:37:31.750 --> 00:37:34.390 align:middle line:84% in a very small population, whatever 00:37:34.390 --> 00:37:38.500 align:middle line:84% we can say about variation is going to disappear. 00:37:38.500 --> 00:37:40.360 align:middle line:90% Now, the contrast of-- 00:37:40.360 --> 00:37:43.100 align:middle line:84% what is opposite of the drift is a mutation. 00:37:43.100 --> 00:37:45.460 align:middle line:84% So we have these random mutations that come up and show 00:37:45.460 --> 00:37:48.850 align:middle line:84% up in our genome, but as you can see, 00:37:48.850 --> 00:37:51.610 align:middle line:84% the funny thing between these mechanism-- that is, 00:37:51.610 --> 00:37:56.710 align:middle line:84% whatever we observe is drift and mutation-- 00:37:56.710 --> 00:37:58.720 align:middle line:84% mutation introduce changes, drift 00:37:58.720 --> 00:38:02.140 align:middle line:84% makes these change stable, so just get rid of them-- 00:38:02.140 --> 00:38:04.960 align:middle line:84% is in sharp contrast with what we 00:38:04.960 --> 00:38:07.870 align:middle line:84% have been taught in high school, that is, 00:38:07.870 --> 00:38:09.400 align:middle line:90% there is some kind of selection. 00:38:09.400 --> 00:38:11.080 align:middle line:90% What is selection here? 00:38:11.080 --> 00:38:14.230 align:middle line:84% Where people get eaten, where people don't run fast enough? 00:38:14.230 --> 00:38:17.600 align:middle line:90% 00:38:17.600 --> 00:38:20.810 align:middle line:84% Is there a way to account for selection in this? 00:38:20.810 --> 00:38:22.490 align:middle line:90% Yes, there is. 00:38:22.490 --> 00:38:26.660 align:middle line:84% So let's keep a quantitative representation of selection. 00:38:26.660 --> 00:38:28.550 align:middle line:84% A quantitative representation of selection 00:38:28.550 --> 00:38:31.270 align:middle line:90% is a function of fitness. 00:38:31.270 --> 00:38:33.650 align:middle line:84% That will tell you how good is somebody 00:38:33.650 --> 00:38:36.590 align:middle line:84% with that particular SNP, with particular allele, 00:38:36.590 --> 00:38:40.970 align:middle line:84% to survive in this particular environment, right? 00:38:40.970 --> 00:38:44.490 align:middle line:84% Now, so suppose you have an allele that has 00:38:44.490 --> 00:38:46.910 align:middle line:90% a distribution of 0.6 and 0.4. 00:38:46.910 --> 00:38:49.610 align:middle line:84% So you apply the Hardy-Weinberg law, 00:38:49.610 --> 00:38:52.100 align:middle line:84% and you have that distribution of homozygous 00:38:52.100 --> 00:38:54.320 align:middle line:84% and the heterozygous in your population. 00:38:54.320 --> 00:38:56.500 align:middle line:90% Then you have a function. 00:38:56.500 --> 00:39:02.570 align:middle line:84% We have a fitness function that will tell you that you have a-- 00:39:02.570 --> 00:39:07.430 align:middle line:84% your selection rate will be 0.2 because you're 00:39:07.430 --> 00:39:11.360 align:middle line:84% going to say that if you have the major allele 00:39:11.360 --> 00:39:13.985 align:middle line:84% or the heterozygous, you are all right. 00:39:13.985 --> 00:39:17.360 align:middle line:84% Your probability of survival is 1. 00:39:17.360 --> 00:39:20.540 align:middle line:84% If you are homozygous at the minor allele, 00:39:20.540 --> 00:39:23.490 align:middle line:84% you have a diminished capacity to survive. 00:39:23.490 --> 00:39:25.970 align:middle line:84% So we can compute out of this fitness function 00:39:25.970 --> 00:39:28.760 align:middle line:84% a selection function of 0.2 that will tell us 00:39:28.760 --> 00:39:31.850 align:middle line:84% how many people we lose in a particular category 00:39:31.850 --> 00:39:33.470 align:middle line:90% at each generation. 00:39:33.470 --> 00:39:36.005 align:middle line:84% Now, if you look at the effect on the first generation, 00:39:36.005 --> 00:39:41.900 align:middle line:84% you see that there is an increase in the homozygous 00:39:41.900 --> 00:39:45.200 align:middle line:84% at the major allele, a decrease in the homozygous-- 00:39:45.200 --> 00:39:49.430 align:middle line:84% in the heterozygous, and sharper decreases in the homozygous 00:39:49.430 --> 00:39:51.050 align:middle line:90% at the minor allele. 00:39:51.050 --> 00:39:53.180 align:middle line:84% But there is something funny here, 00:39:53.180 --> 00:39:56.630 align:middle line:84% that because we used a differential there, 00:39:56.630 --> 00:40:01.550 align:middle line:84% the smaller is the number of individuals 00:40:01.550 --> 00:40:03.980 align:middle line:84% that are homozygous at the minor allele, 00:40:03.980 --> 00:40:08.210 align:middle line:84% the smaller will be their depletion, right? 00:40:08.210 --> 00:40:10.280 align:middle line:84% So this mutation, that's a mutation 00:40:10.280 --> 00:40:14.240 align:middle line:90% doesn't really go away ever. 00:40:14.240 --> 00:40:18.830 align:middle line:84% It will simply stay there and slowly be taken out-- 00:40:18.830 --> 00:40:20.810 align:middle line:84% taking out individuals, but it will 00:40:20.810 --> 00:40:26.030 align:middle line:84% be lurking in the genetic code for the years to come, OK? 00:40:26.030 --> 00:40:28.130 align:middle line:84% We have no explanation in this way 00:40:28.130 --> 00:40:30.830 align:middle line:90% of why dinosaurs disappeared. 00:40:30.830 --> 00:40:33.590 align:middle line:84% This simply tells us that a particular mutation will 00:40:33.590 --> 00:40:35.810 align:middle line:84% keep floating in our population and we 00:40:35.810 --> 00:40:38.210 align:middle line:84% have no good mathematical models to explain 00:40:38.210 --> 00:40:39.900 align:middle line:90% how they will disappear. 00:40:39.900 --> 00:40:43.590 align:middle line:84% Now, that's this-- everything is based on Hardy-Weinberg law, 00:40:43.590 --> 00:40:44.090 align:middle line:90% right? 00:40:44.090 --> 00:40:45.300 align:middle line:90% Does it really work? 00:40:45.300 --> 00:40:49.970 align:middle line:84% So this guy, this [INAUDIBLE] in 1975, in England, 00:40:49.970 --> 00:40:54.380 align:middle line:90% they made this experiment. 00:40:54.380 --> 00:40:58.010 align:middle line:90% They took blood samples. 00:40:58.010 --> 00:41:01.340 align:middle line:84% This is their blood groups, which are, again, 00:41:01.340 --> 00:41:04.280 align:middle line:90% governed by one single allele. 00:41:04.280 --> 00:41:08.570 align:middle line:84% And they look at the population of this group, 00:41:08.570 --> 00:41:10.850 align:middle line:84% this random population, and they found 00:41:10.850 --> 00:41:15.650 align:middle line:84% that if you compute the expected values using Hardy-Weinberg 00:41:15.650 --> 00:41:20.570 align:middle line:84% and you compute the values, the observed values, 00:41:20.570 --> 00:41:24.170 align:middle line:84% these numbers are remarkably similar. 00:41:24.170 --> 00:41:27.410 align:middle line:84% You see, so we would expect 261.54 00:41:27.410 --> 00:41:30.320 align:middle line:90% to be Mm from Hardy-Weinberg. 00:41:30.320 --> 00:41:33.710 align:middle line:90% We get 363. 00:41:33.710 --> 00:41:38.960 align:middle line:84% We get 636 for Mn, and we get 334. 00:41:38.960 --> 00:41:42.170 align:middle line:84% Now, these numbers are so similar that somebody wrote 00:41:42.170 --> 00:41:45.830 align:middle line:84% a paper showing that actually they had been falsified, 00:41:45.830 --> 00:41:48.150 align:middle line:90% because you don't-- 00:41:48.150 --> 00:41:51.290 align:middle line:84% the precision of this number is above 00:41:51.290 --> 00:41:54.290 align:middle line:84% any reasonable statistical expectation of precision 00:41:54.290 --> 00:41:59.980 align:middle line:90% for a particular law, OK? 00:41:59.980 --> 00:42:01.672 align:middle line:84% But let's say they falsified the number 00:42:01.672 --> 00:42:03.130 align:middle line:84% because they didn't know statistics 00:42:03.130 --> 00:42:05.320 align:middle line:84% and they didn't know they could get a way with the better 00:42:05.320 --> 00:42:05.900 align:middle line:90% results. 00:42:05.900 --> 00:42:07.358 align:middle line:84% But let's buy this thing because we 00:42:07.358 --> 00:42:10.210 align:middle line:84% have seen in other population that it's not really that close 00:42:10.210 --> 00:42:11.710 align:middle line:90% but it really works. 00:42:11.710 --> 00:42:13.615 align:middle line:84% But we've also seen some population in which 00:42:13.615 --> 00:42:15.250 align:middle line:90% it doesn't really work. 00:42:15.250 --> 00:42:18.640 align:middle line:84% So beta model being sickle cell anemia 00:42:18.640 --> 00:42:21.340 align:middle line:90% in West Africa, if you look-- 00:42:21.340 --> 00:42:26.200 align:middle line:84% if you compute the distribution using Hardy-Weinberg, 00:42:26.200 --> 00:42:28.700 align:middle line:84% and you compute and you observe the distribution 00:42:28.700 --> 00:42:31.120 align:middle line:84% in the population, you have this remarkable change. 00:42:31.120 --> 00:42:34.380 align:middle line:90% 00:42:34.380 --> 00:42:36.720 align:middle line:84% Look at the difference of those two numbers. 00:42:36.720 --> 00:42:41.320 align:middle line:84% We expect that, to have 254, and as a matter of fact, 00:42:41.320 --> 00:42:43.185 align:middle line:90% we have 64. 00:42:43.185 --> 00:42:46.910 align:middle line:90% 00:42:46.910 --> 00:42:47.967 align:middle line:90% Why? 00:42:47.967 --> 00:42:49.252 align:middle line:90% AUDIENCE: [INAUDIBLE]. 00:42:49.252 --> 00:42:50.252 align:middle line:90% MARCO RAMONI: Say again. 00:42:50.252 --> 00:42:53.447 align:middle line:84% AUDIENCE: [INAUDIBLE],, so you can't observe. 00:42:53.447 --> 00:42:54.155 align:middle line:90% MARCO RAMONI: OK. 00:42:54.155 --> 00:43:05.570 align:middle line:90% 00:43:05.570 --> 00:43:07.080 align:middle line:84% I think I-- no, I made a mistake. 00:43:07.080 --> 00:43:09.110 align:middle line:90% It's flipped. 00:43:09.110 --> 00:43:12.320 align:middle line:84% So we have-- sorry, we have 600 and something, not 64. 00:43:12.320 --> 00:43:15.672 align:middle line:84% This is 645, so your explanation is the opposite. 00:43:15.672 --> 00:43:17.630 align:middle line:84% So you have to explain the opposite phenomenon, 00:43:17.630 --> 00:43:19.917 align:middle line:84% not why it's smaller but because it's bigger. 00:43:19.917 --> 00:43:20.888 align:middle line:90% [LAUGHTER] 00:43:20.888 --> 00:43:22.430 align:middle line:84% As a matter of fact, you have a-- you 00:43:22.430 --> 00:43:25.130 align:middle line:84% see you have more homozygous-- heterozygous in the middle? 00:43:25.130 --> 00:43:28.800 align:middle line:90% 00:43:28.800 --> 00:43:29.430 align:middle line:90% Oh, no, no. 00:43:29.430 --> 00:43:30.090 align:middle line:90% The number is correct. 00:43:30.090 --> 00:43:30.945 align:middle line:90% So this is the thing. 00:43:30.945 --> 00:43:31.445 align:middle line:90% [LAUGHTER] 00:43:31.445 --> 00:43:35.010 align:middle line:84% The inconsistent thing is that you have too many heterozygous, 00:43:35.010 --> 00:43:38.250 align:middle line:84% given the homozygous and the minor alleles. 00:43:38.250 --> 00:43:40.890 align:middle line:90% You shouldn't have them. 00:43:40.890 --> 00:43:41.895 align:middle line:90% You see? 00:43:41.895 --> 00:43:44.596 align:middle line:84% AUDIENCE: But isn't that evidence of a selection, 00:43:44.596 --> 00:43:47.010 align:middle line:90% evidence of gene selection? 00:43:47.010 --> 00:43:50.280 align:middle line:84% MARCO RAMONI: You get two pizzas. 00:43:50.280 --> 00:43:54.250 align:middle line:84% Because the heterozygous get protection against malaria, 00:43:54.250 --> 00:43:57.630 align:middle line:84% so in doing so, you have a bias towards having 00:43:57.630 --> 00:43:59.730 align:middle line:84% a lot of heterozygous in the middle 00:43:59.730 --> 00:44:02.520 align:middle line:84% and keep the allele around in the population. 00:44:02.520 --> 00:44:05.940 align:middle line:84% And the other below there doesn't remain-- 00:44:05.940 --> 00:44:08.970 align:middle line:84% doesn't really change, but the selective advantage 00:44:08.970 --> 00:44:13.350 align:middle line:84% is for these 5,400 people who have the heterozygous. 00:44:13.350 --> 00:44:15.420 align:middle line:84% So in that particular case-- this 00:44:15.420 --> 00:44:18.060 align:middle line:84% is something that probably wouldn't be a good test 00:44:18.060 --> 00:44:20.400 align:middle line:84% to make here, because here malaria is not really 00:44:20.400 --> 00:44:21.190 align:middle line:90% a big issue. 00:44:21.190 --> 00:44:24.030 align:middle line:84% But in West Africa it's going to be a big problem. 00:44:24.030 --> 00:44:26.790 align:middle line:84% AUDIENCE: If we did the-- did that test 00:44:26.790 --> 00:44:28.500 align:middle line:84% at birth or before birth, would it 00:44:28.500 --> 00:44:31.255 align:middle line:84% be consistent with the prediction? 00:44:31.255 --> 00:44:32.380 align:middle line:90% MARCO RAMONI: Before birth? 00:44:32.380 --> 00:44:35.070 align:middle line:84% AUDIENCE: If you did if you did the same distribution 00:44:35.070 --> 00:44:38.262 align:middle line:84% of the alleles before they had this pressure of malaria? 00:44:38.262 --> 00:44:39.720 align:middle line:84% MARCO RAMONI: Well, that's is the-- 00:44:39.720 --> 00:44:42.120 align:middle line:84% AUDIENCE: Is it the dying-off factor that's causing 00:44:42.120 --> 00:44:43.328 align:middle line:90% the difference or something-- 00:44:43.328 --> 00:44:45.000 align:middle line:84% MARCO RAMONI: No, there is no dying. 00:44:45.000 --> 00:44:46.800 align:middle line:90% Well, so let's put it this way. 00:44:46.800 --> 00:44:49.020 align:middle line:84% We actually do it before they're born, 00:44:49.020 --> 00:44:52.230 align:middle line:84% because the selection is actually on this parents. 00:44:52.230 --> 00:44:54.730 align:middle line:84% The pressure, as usual, is on the parents. 00:44:54.730 --> 00:44:57.240 align:middle line:84% So if you keep these things around, 00:44:57.240 --> 00:45:02.010 align:middle line:84% the chances that these people will go in older age 00:45:02.010 --> 00:45:04.260 align:middle line:90% and reproduce are higher. 00:45:04.260 --> 00:45:09.390 align:middle line:84% So we-- the entire population has a global advantage 00:45:09.390 --> 00:45:16.500 align:middle line:84% in keeping around this particular variation, 00:45:16.500 --> 00:45:19.980 align:middle line:84% but only in the particular form, which is the heterozygous. 00:45:19.980 --> 00:45:22.810 align:middle line:84% The homozygous is going to be too bad for you, 00:45:22.810 --> 00:45:24.840 align:middle line:84% but the heterozygous is going to strike 00:45:24.840 --> 00:45:27.000 align:middle line:84% a nice balance between getting the disease 00:45:27.000 --> 00:45:30.780 align:middle line:84% and actually being protected by-- 00:45:30.780 --> 00:45:32.130 align:middle line:90% against malaria. 00:45:32.130 --> 00:45:33.588 align:middle line:84% If you do something like this here, 00:45:33.588 --> 00:45:35.880 align:middle line:84% you change the environmental condition, you're screwed. 00:45:35.880 --> 00:45:40.980 align:middle line:84% Here, I'm pretty sure, if you do the same test, 00:45:40.980 --> 00:45:44.670 align:middle line:84% this will not show up because the population as a whole 00:45:44.670 --> 00:45:50.790 align:middle line:84% doesn't have a pressure for this. 00:45:50.790 --> 00:45:53.580 align:middle line:90% Now, how do we map these things? 00:45:53.580 --> 00:45:56.910 align:middle line:90% How do we use a marker as such? 00:45:56.910 --> 00:46:04.260 align:middle line:84% Well, first we want to find the genetic basis, a genotype 00:46:04.260 --> 00:46:08.370 align:middle line:84% for a particular disease that is our phenotype, 00:46:08.370 --> 00:46:14.520 align:middle line:84% and what's the mindset for somebody looking for this? 00:46:14.520 --> 00:46:17.730 align:middle line:84% We are not really looking for the SNP or the mutation that 00:46:17.730 --> 00:46:18.600 align:middle line:90% is changing there. 00:46:18.600 --> 00:46:22.230 align:middle line:84% We are trying to find the marker that is in the proximity-- 00:46:22.230 --> 00:46:24.390 align:middle line:84% well, we'll be happy to find the marker that 00:46:24.390 --> 00:46:28.980 align:middle line:84% is in the proximity of the actual codes. 00:46:28.980 --> 00:46:32.250 align:middle line:84% So the mindset is more or less this one. 00:46:32.250 --> 00:46:35.370 align:middle line:84% I have a marker, and I observe in my data 00:46:35.370 --> 00:46:39.360 align:middle line:84% set a dependency between my phenotype and my marker. 00:46:39.360 --> 00:46:42.570 align:middle line:84% But in reality, the extent of my claim cannot be, 00:46:42.570 --> 00:46:44.760 align:middle line:84% this marker cause this phenotype. 00:46:44.760 --> 00:46:47.370 align:middle line:84% The extent of my claim can only be, there-- 00:46:47.370 --> 00:46:49.560 align:middle line:84% this marker is linkaged to equilibrium 00:46:49.560 --> 00:46:51.750 align:middle line:84% with some real genotype that I haven't 00:46:51.750 --> 00:46:55.470 align:middle line:84% observed that is actually causing the phenotype. 00:46:55.470 --> 00:46:59.670 align:middle line:84% And you see where the complexities start emerging. 00:46:59.670 --> 00:47:03.810 align:middle line:84% The first complexity is that the phenotype and-- the genotype 00:47:03.810 --> 00:47:07.260 align:middle line:84% and the phenotype may have a complex form of causality. 00:47:07.260 --> 00:47:09.690 align:middle line:84% They may have an interaction with 00:47:09.690 --> 00:47:13.260 align:middle line:84% an environmental condition, or say they 00:47:13.260 --> 00:47:15.600 align:middle line:90% have incomplete penetrance. 00:47:15.600 --> 00:47:17.880 align:middle line:84% And the other thing is that, our group 00:47:17.880 --> 00:47:19.980 align:middle line:84% is the linkaged equilibrium between the marker 00:47:19.980 --> 00:47:21.570 align:middle line:90% and genotype. 00:47:21.570 --> 00:47:25.920 align:middle line:84% I don't expect to have everybody getting exactly the same pair 00:47:25.920 --> 00:47:29.100 align:middle line:90% of genotypes, right? 00:47:29.100 --> 00:47:34.290 align:middle line:84% So this is where stochasticity comes in, 00:47:34.290 --> 00:47:37.170 align:middle line:84% and this is what is making our life slightly more 00:47:37.170 --> 00:47:40.680 align:middle line:84% complex but definitely more interesting. 00:47:40.680 --> 00:47:45.060 align:middle line:84% Now, traits don't really follow single gene models. 00:47:45.060 --> 00:47:47.970 align:middle line:84% If you look at the list of diseases they gave, 00:47:47.970 --> 00:47:50.370 align:middle line:84% they are kind of minor diseases, minor. 00:47:50.370 --> 00:47:51.660 align:middle line:90% They have small incidence. 00:47:51.660 --> 00:47:54.540 align:middle line:84% They are horrible diseases, but they are not 00:47:54.540 --> 00:48:00.420 align:middle line:84% going to save the 50% of people who die of stroke or people 00:48:00.420 --> 00:48:02.760 align:middle line:84% who have diabetes or people that have this 00:48:02.760 --> 00:48:08.040 align:middle line:84% for more complicated alleles factors 00:48:08.040 --> 00:48:11.770 align:middle line:84% that cause more interesting diseases. 00:48:11.770 --> 00:48:15.210 align:middle line:84% On the other hand, even some Mendelian traits 00:48:15.210 --> 00:48:18.600 align:middle line:90% are actually complex. 00:48:18.600 --> 00:48:21.700 align:middle line:90% So if you look at the-- 00:48:21.700 --> 00:48:25.890 align:middle line:84% so the-- and sickle cell anemia is a classical Mendelian 00:48:25.890 --> 00:48:27.620 align:middle line:90% disease, but-- 00:48:27.620 --> 00:48:30.320 align:middle line:90% so we know why you have it. 00:48:30.320 --> 00:48:35.000 align:middle line:84% But the phenotypic variability of the disease is immense. 00:48:35.000 --> 00:48:40.760 align:middle line:84% There are children that die at 13 by stroke, 00:48:40.760 --> 00:48:43.275 align:middle line:84% and there are people that live forever. 00:48:43.275 --> 00:48:44.900 align:middle line:84% And there are perfectly healthy people. 00:48:44.900 --> 00:48:46.192 align:middle line:90% We don't know why this happens. 00:48:46.192 --> 00:48:47.930 align:middle line:84% We know these people get the disease, 00:48:47.930 --> 00:48:50.520 align:middle line:84% but we don't know what the-- why this happens. 00:48:50.520 --> 00:48:55.160 align:middle line:84% And so people are, for instance, studying the early mortality 00:48:55.160 --> 00:49:01.780 align:middle line:84% of kids with sickle cell anemia, try to find [INAUDIBLE] SNP, 00:49:01.780 --> 00:49:03.530 align:middle line:84% and they say, well, you know, these people 00:49:03.530 --> 00:49:07.370 align:middle line:84% get this particular disease that is genetically found. 00:49:07.370 --> 00:49:11.630 align:middle line:84% Maybe the difference between their long-term survival-- 00:49:11.630 --> 00:49:13.550 align:middle line:84% the chance of their long-term survival 00:49:13.550 --> 00:49:18.250 align:middle line:84% is really due exclusively to another SNP. 00:49:18.250 --> 00:49:21.260 align:middle line:84% Actually, they are almost finding it. 00:49:21.260 --> 00:49:23.750 align:middle line:84% Another problem is incomplete penetrance 00:49:23.750 --> 00:49:25.980 align:middle line:84% when you have an interaction with another variable. 00:49:25.980 --> 00:49:28.460 align:middle line:84% This is a recent, very famous case called Bcr1. 00:49:28.460 --> 00:49:34.220 align:middle line:84% Bcr1 is a gene, is a particular locus 00:49:34.220 --> 00:49:40.280 align:middle line:84% that will be a predictor of the greater risk of developing 00:49:40.280 --> 00:49:41.810 align:middle line:90% breast cancer in a woman. 00:49:41.810 --> 00:49:44.360 align:middle line:84% Now, we know that it's very difficult for a woman 00:49:44.360 --> 00:49:47.220 align:middle line:84% to develop breast cancer before menopause, 00:49:47.220 --> 00:49:51.080 align:middle line:84% so you expect in their 40s to have a very low incidence. 00:49:51.080 --> 00:49:56.330 align:middle line:84% But if look at-- if you average all this, the population in-- 00:49:56.330 --> 00:50:00.710 align:middle line:84% across people with BCR 1 and people without Bcr1, 00:50:00.710 --> 00:50:04.340 align:middle line:84% you will see that the evidence is not really that big. 00:50:04.340 --> 00:50:07.670 align:middle line:84% They get the slight increase in chances of developing 00:50:07.670 --> 00:50:09.060 align:middle line:90% the breast cancer. 00:50:09.060 --> 00:50:11.900 align:middle line:84% But if you split them according to their age, 00:50:11.900 --> 00:50:15.710 align:middle line:84% so you use age as an environmental factor, 00:50:15.710 --> 00:50:20.760 align:middle line:84% you see how the changes increase along time. 00:50:20.760 --> 00:50:24.740 align:middle line:84% So when you're 40, it's 37 chances-- 00:50:24.740 --> 00:50:26.390 align:middle line:90% 37% chances. 00:50:26.390 --> 00:50:30.560 align:middle line:84% But when you're 80, it's 85% chances 00:50:30.560 --> 00:50:32.100 align:middle line:84% that it's going to make a change. 00:50:32.100 --> 00:50:34.353 align:middle line:84% But if you put them all together, 00:50:34.353 --> 00:50:38.130 align:middle line:84% eh, you have something that is not such a clear result. 00:50:38.130 --> 00:50:40.580 align:middle line:84% On the other hand, the challenge here 00:50:40.580 --> 00:50:44.360 align:middle line:84% is actually to find out which is the environmental factor that 00:50:44.360 --> 00:50:45.650 align:middle line:90% is actually doing the deed. 00:50:45.650 --> 00:50:47.330 align:middle line:84% This way is to find out that it's 00:50:47.330 --> 00:50:49.862 align:middle line:84% age that is splitting these women 00:50:49.862 --> 00:50:50.945 align:middle line:90% in these different groups. 00:50:50.945 --> 00:50:55.080 align:middle line:90% 00:50:55.080 --> 00:50:58.070 align:middle line:84% So this is a necessary, sufficient problem, 00:50:58.070 --> 00:51:00.710 align:middle line:84% but we have also redundancy problem. 00:51:00.710 --> 00:51:02.870 align:middle line:84% For instance, we have retinitis pigmentosa 00:51:02.870 --> 00:51:07.440 align:middle line:84% that may be caused by 14 independent SNPs, mutations, 00:51:07.440 --> 00:51:07.940 align:middle line:90% right? 00:51:07.940 --> 00:51:10.730 align:middle line:84% So any of these mutations will show up 00:51:10.730 --> 00:51:14.900 align:middle line:84% positive to a particular genetic test for association 00:51:14.900 --> 00:51:18.620 align:middle line:90% with the disease. 00:51:18.620 --> 00:51:22.520 align:middle line:84% And we have other diseases that are 00:51:22.520 --> 00:51:24.260 align:middle line:90% known to have polygenic causes. 00:51:24.260 --> 00:51:25.970 align:middle line:84% If you have only one of these, you 00:51:25.970 --> 00:51:28.910 align:middle line:84% will not develop the disease, like the Hirschsprung disease, 00:51:28.910 --> 00:51:33.020 align:middle line:84% in which you'd need two different mutations on two 00:51:33.020 --> 00:51:35.360 align:middle line:84% different chromosomes, which, by the way, 00:51:35.360 --> 00:51:38.720 align:middle line:84% opens an interesting question on, 00:51:38.720 --> 00:51:42.140 align:middle line:84% what is the relationship, the evolutionary relationship, 00:51:42.140 --> 00:51:45.260 align:middle line:84% of objects that leave off loci that are 00:51:45.260 --> 00:51:49.010 align:middle line:90% on two different chromosomes? 00:51:49.010 --> 00:51:51.590 align:middle line:90% They are completely independent? 00:51:51.590 --> 00:51:53.540 align:middle line:84% Well, we have very complex systems 00:51:53.540 --> 00:51:56.960 align:middle line:84% that are displaced around-- there are genes displaced 00:51:56.960 --> 00:52:00.950 align:middle line:84% around different chromosomes, and we 00:52:00.950 --> 00:52:04.370 align:middle line:84% are able to rebuild this system from one generation to another. 00:52:04.370 --> 00:52:07.430 align:middle line:90% 00:52:07.430 --> 00:52:10.750 align:middle line:84% How do we combine this with the idea that if you measure things 00:52:10.750 --> 00:52:14.290 align:middle line:84% by centimorgan, two things-- it doesn't make any sense 00:52:14.290 --> 00:52:15.370 align:middle line:90% to have-- 00:52:15.370 --> 00:52:17.440 align:middle line:84% to compare the distance of two loci 00:52:17.440 --> 00:52:19.790 align:middle line:90% on two different chromosomes? 00:52:19.790 --> 00:52:22.660 align:middle line:84% So this is a notion, kind of radical notion, 00:52:22.660 --> 00:52:27.460 align:middle line:84% but proposed in the '30s by Sir Ronald Fisher, father 00:52:27.460 --> 00:52:30.520 align:middle line:84% of statistics and modern genetics, 00:52:30.520 --> 00:52:34.480 align:middle line:90% under the name of affinity. 00:52:34.480 --> 00:52:36.400 align:middle line:84% He didn't have genotypes at that time, 00:52:36.400 --> 00:52:39.700 align:middle line:84% but it was a paper on nature, talking about this very strange 00:52:39.700 --> 00:52:42.760 align:middle line:84% phenomenon and how likely it was that there was really 00:52:42.760 --> 00:52:44.530 align:middle line:84% no relationship on things that were 00:52:44.530 --> 00:52:49.910 align:middle line:84% very far on the genetic code, in this code he was imagining. 00:52:49.910 --> 00:52:53.035 align:middle line:84% And they were, at the same time, functionally tightly related. 00:52:53.035 --> 00:52:56.480 align:middle line:90% 00:52:56.480 --> 00:53:01.930 align:middle line:84% So how do we-- how we try to dissect all this complex stuff 00:53:01.930 --> 00:53:03.190 align:middle line:90% and also the simple one? 00:53:03.190 --> 00:53:04.670 align:middle line:84% Well, you know, the traditional way 00:53:04.670 --> 00:53:10.330 align:middle line:84% is to do the same game we did for Victoria's secrets, right, 00:53:10.330 --> 00:53:15.160 align:middle line:84% find a large pedigree, wait, you know, those 150, 200 years, 00:53:15.160 --> 00:53:18.340 align:middle line:84% and see what happens across three generations. 00:53:18.340 --> 00:53:20.780 align:middle line:84% We can do this with the drosophila much faster, 00:53:20.780 --> 00:53:23.380 align:middle line:84% but if we want to approach human diseases, 00:53:23.380 --> 00:53:25.850 align:middle line:84% drosophila doesn't really always cut it. 00:53:25.850 --> 00:53:29.750 align:middle line:84% So if we want to study the same thing in humans, what do we do? 00:53:29.750 --> 00:53:32.290 align:middle line:90% One is to find a large pedigree. 00:53:32.290 --> 00:53:34.330 align:middle line:84% One is to say, well, you know, I don't really 00:53:34.330 --> 00:53:35.590 align:middle line:90% need a large pedigree. 00:53:35.590 --> 00:53:37.690 align:middle line:90% What I can get are trios. 00:53:37.690 --> 00:53:41.950 align:middle line:84% I get mother, father, and child, a very modern nuclear family. 00:53:41.950 --> 00:53:44.770 align:middle line:84% And then I imagine-- because I know the parents-- 00:53:44.770 --> 00:53:48.820 align:middle line:84% I imagine that I will get statistical evidence 00:53:48.820 --> 00:53:51.790 align:middle line:84% of transmission by repeating the same measurement 00:53:51.790 --> 00:53:53.960 align:middle line:90% in different spheres. 00:53:53.960 --> 00:53:54.460 align:middle line:90% Right? 00:53:54.460 --> 00:53:58.370 align:middle line:84% Like, they were a huge pedigree, and in this case, 00:53:58.370 --> 00:54:01.060 align:middle line:84% I can do everything in two generations. 00:54:01.060 --> 00:54:03.790 align:middle line:90% 00:54:03.790 --> 00:54:06.820 align:middle line:84% I can have-- but sometimes, for some diseases, 00:54:06.820 --> 00:54:09.280 align:middle line:84% finding parents is very difficult. 00:54:09.280 --> 00:54:12.190 align:middle line:84% If you're looking for two complex traits that 00:54:12.190 --> 00:54:14.745 align:middle line:84% show up late in age or-- there are 00:54:14.745 --> 00:54:16.120 align:middle line:84% people that are actually studying 00:54:16.120 --> 00:54:19.090 align:middle line:90% this, the basis of longevity. 00:54:19.090 --> 00:54:23.380 align:middle line:84% And they recruit people in their 90s, late 90s, early first 00:54:23.380 --> 00:54:25.130 align:middle line:90% century-- 00:54:25.130 --> 00:54:26.475 align:middle line:90% early second century. 00:54:26.475 --> 00:54:28.600 align:middle line:84% It's kind of difficult to track down their parents, 00:54:28.600 --> 00:54:32.810 align:middle line:84% and most of all, it's difficult to genotype them. 00:54:32.810 --> 00:54:35.620 align:middle line:90% So what do you do in this case? 00:54:35.620 --> 00:54:37.150 align:middle line:90% You get brothers. 00:54:37.150 --> 00:54:39.790 align:middle line:84% If you get the brother, you try to figure out 00:54:39.790 --> 00:54:42.210 align:middle line:90% what was the genotype. 00:54:42.210 --> 00:54:44.710 align:middle line:84% So you assume that there is some kind of random distribution 00:54:44.710 --> 00:54:46.930 align:middle line:84% of these things across the two children, 00:54:46.930 --> 00:54:50.230 align:middle line:84% and you try to figure out what was the original genotype 00:54:50.230 --> 00:54:53.140 align:middle line:90% of the parents. 00:54:53.140 --> 00:54:57.970 align:middle line:84% Or you can do a standard case control experiment in which 00:54:57.970 --> 00:55:01.790 align:middle line:90% you collect a bunch of people-- 00:55:01.790 --> 00:55:04.430 align:middle line:84% half of them or both of them have the disease, 00:55:04.430 --> 00:55:06.010 align:middle line:84% other then don't have the disease-- 00:55:06.010 --> 00:55:13.450 align:middle line:84% and compare the association between a SNP, a mutation, 00:55:13.450 --> 00:55:16.780 align:middle line:90% and if so, all three. 00:55:16.780 --> 00:55:20.470 align:middle line:84% There is another way to categorize these experiments. 00:55:20.470 --> 00:55:24.160 align:middle line:84% One is-- one are the experiments that are, like, double-sided, 00:55:24.160 --> 00:55:25.690 align:middle line:90% like case control experiments. 00:55:25.690 --> 00:55:29.290 align:middle line:84% I have two-- both sides of the story, the affected and the not 00:55:29.290 --> 00:55:30.160 align:middle line:90% affected. 00:55:30.160 --> 00:55:33.550 align:middle line:84% Or I can have single-sided studies-- 00:55:33.550 --> 00:55:35.620 align:middle line:84% and I will show you how you analyze them-- 00:55:35.620 --> 00:55:38.080 align:middle line:90% in which you recruit parents-- 00:55:38.080 --> 00:55:41.500 align:middle line:84% they are healthy parents with an affected child-- 00:55:41.500 --> 00:55:44.380 align:middle line:84% and you try to see what is the transmission that 00:55:44.380 --> 00:55:47.980 align:middle line:84% goes from parents, child, that make the child affected. 00:55:47.980 --> 00:55:50.980 align:middle line:84% In this case, all your pool, all your recruited pool, 00:55:50.980 --> 00:55:54.610 align:middle line:84% would be made by people that are either affected or are 00:55:54.610 --> 00:55:55.990 align:middle line:90% related to an affected person. 00:55:55.990 --> 00:55:57.460 align:middle line:84% Which actually makes, usually, life 00:55:57.460 --> 00:56:00.980 align:middle line:84% much easier for the recruiter because, if I have to give-- 00:56:00.980 --> 00:56:04.480 align:middle line:84% to volunteer for study if I am affected 00:56:04.480 --> 00:56:06.760 align:middle line:84% or my child is affected, I am going 00:56:06.760 --> 00:56:09.190 align:middle line:84% to be more willing to volunteer for the study 00:56:09.190 --> 00:56:12.340 align:middle line:84% than if I'm just a random bozo that doesn't have the disease 00:56:12.340 --> 00:56:15.307 align:middle line:84% and by design I have to be independent of any subject 00:56:15.307 --> 00:56:16.015 align:middle line:90% with the disease. 00:56:16.015 --> 00:56:20.000 align:middle line:90% 00:56:20.000 --> 00:56:26.530 align:middle line:84% So I will go into details for each of these studies. 00:56:26.530 --> 00:56:30.790 align:middle line:84% So the first thing we can do is to do linkage analysis. 00:56:30.790 --> 00:56:34.710 align:middle line:84% That is a traditional analysis of pedigrees. 00:56:34.710 --> 00:56:38.280 align:middle line:84% The second way we can do-- we can go is to-- 00:56:38.280 --> 00:56:39.245 align:middle line:90% is allele sharing. 00:56:39.245 --> 00:56:41.570 align:middle line:84% Do you remember what I was telling you about, 00:56:41.570 --> 00:56:45.340 align:middle line:84% about brothers, and siblings in general? 00:56:45.340 --> 00:56:48.180 align:middle line:84% So if you have siblings, what you can do 00:56:48.180 --> 00:56:50.490 align:middle line:84% is, imagine that there is a random distribution 00:56:50.490 --> 00:56:53.970 align:middle line:84% for a particular SNP, for a particular mutation. 00:56:53.970 --> 00:56:56.580 align:middle line:84% And you would expect that if you have two brothers, 00:56:56.580 --> 00:56:59.670 align:middle line:84% these two brothers will have a particular probability 00:56:59.670 --> 00:57:01.540 align:middle line:84% of getting one allele or another. 00:57:01.540 --> 00:57:05.160 align:middle line:84% Now, if you start deviating from this distribution, 00:57:05.160 --> 00:57:09.090 align:middle line:84% then you start becoming a bit suspicious. 00:57:09.090 --> 00:57:12.480 align:middle line:84% The association studies are standard case control studies 00:57:12.480 --> 00:57:14.340 align:middle line:84% in which you compute the likelihood 00:57:14.340 --> 00:57:17.190 align:middle line:84% that a particular SNP is causing, 00:57:17.190 --> 00:57:19.200 align:middle line:90% affecting your outcome. 00:57:19.200 --> 00:57:21.240 align:middle line:84% TDT, Transmission Disequilibrium Test, 00:57:21.240 --> 00:57:25.380 align:middle line:84% is the thing we actually use for the trios, 00:57:25.380 --> 00:57:27.400 align:middle line:90% and I will tell you in a second. 00:57:27.400 --> 00:57:34.140 align:middle line:84% And then for complex traits, like quantitative traits, 00:57:34.140 --> 00:57:36.300 align:middle line:84% things that are not really binary diseases 00:57:36.300 --> 00:57:41.310 align:middle line:84% but are, for instance, your resistance to-- 00:57:41.310 --> 00:57:43.210 align:middle line:84% your inclination to start drinking, 00:57:43.210 --> 00:57:44.820 align:middle line:84% which may have different degrees, 00:57:44.820 --> 00:57:48.450 align:middle line:84% or your inclination to quit smoking, which, again, 00:57:48.450 --> 00:57:49.755 align:middle line:90% can have different degrees. 00:57:49.755 --> 00:57:52.810 align:middle line:90% 00:57:52.810 --> 00:57:54.630 align:middle line:84% The best way of doing them, of looking 00:57:54.630 --> 00:57:59.790 align:middle line:84% at these quantitative traits, is to use animal models, so 00:57:59.790 --> 00:58:02.430 align:middle line:84% crosses, which is not something we are allowed to do in humans. 00:58:02.430 --> 00:58:05.730 align:middle line:90% 00:58:05.730 --> 00:58:08.490 align:middle line:84% Typically, these collections are hypothesis driven. 00:58:08.490 --> 00:58:12.300 align:middle line:84% So what these people do is, they wake up in the morning, 00:58:12.300 --> 00:58:15.510 align:middle line:84% they read a lot of papers the night before, 00:58:15.510 --> 00:58:17.790 align:middle line:84% and they take a shower and they say, 00:58:17.790 --> 00:58:23.410 align:middle line:84% oh, here are seven maybe related to my disease. 00:58:23.410 --> 00:58:25.800 align:middle line:90% So they go out, they ask money. 00:58:25.800 --> 00:58:29.520 align:middle line:84% They recruit a bunch of people, use the money 00:58:29.520 --> 00:58:32.760 align:middle line:84% to genotype these people, run an experiment, 00:58:32.760 --> 00:58:35.820 align:middle line:84% and write a paper in which they either find something 00:58:35.820 --> 00:58:37.470 align:middle line:90% or they find nothing. 00:58:37.470 --> 00:58:39.270 align:middle line:84% Usually, if you don't find nothing, 00:58:39.270 --> 00:58:42.870 align:middle line:84% you can write a paper, and this induced misbehavior 00:58:42.870 --> 00:58:44.790 align:middle line:90% in scientists. 00:58:44.790 --> 00:58:51.210 align:middle line:84% The challenge here is that doing this is the old-fashioned way. 00:58:51.210 --> 00:58:55.380 align:middle line:84% What is very precious for us today, or in this case, 00:58:55.380 --> 00:58:57.310 align:middle line:90% is the samples of people. 00:58:57.310 --> 00:59:00.180 align:middle line:84% Say, imagine you have a single drop of sample-- or of a-- 00:59:00.180 --> 00:59:03.840 align:middle line:84% from an individual and you can do a gazillions of these SNPs, 00:59:03.840 --> 00:59:07.080 align:middle line:84% but in this case, you don't have to read a lot of papers. 00:59:07.080 --> 00:59:08.863 align:middle line:90% What you can do is-- 00:59:08.863 --> 00:59:10.030 align:middle line:90% or you can have your doubts. 00:59:10.030 --> 00:59:11.610 align:middle line:84% You can say, well, maybe it's not only this one. 00:59:11.610 --> 00:59:12.300 align:middle line:90% Maybe it's the second. 00:59:12.300 --> 00:59:12.967 align:middle line:90% Maybe the third. 00:59:12.967 --> 00:59:13.890 align:middle line:90% Maybe it's the fifth. 00:59:13.890 --> 00:59:18.990 align:middle line:84% What if I collect 500 SNPs and then test all of them? 00:59:18.990 --> 00:59:22.350 align:middle line:90% What if I do an analysis that-- 00:59:22.350 --> 00:59:23.610 align:middle line:90% we don't collect any SNP? 00:59:23.610 --> 00:59:26.400 align:middle line:84% We collect all of them and try to look, 00:59:26.400 --> 00:59:32.022 align:middle line:84% to fish for dependencies in the data set. 00:59:32.022 --> 00:59:34.740 align:middle line:84% OK, in this case, it becomes really, really difficult. 00:59:34.740 --> 00:59:38.160 align:middle line:84% This is how things used to be done. 00:59:38.160 --> 00:59:39.780 align:middle line:90% And this is a three-generation-- 00:59:39.780 --> 00:59:44.170 align:middle line:84% 3, 4-- four-generation linkage study. 00:59:44.170 --> 00:59:48.182 align:middle line:84% And in this case, you have a particular disease, and people 00:59:48.182 --> 00:59:49.890 align:middle line:84% that are affected, they are the red ones. 00:59:49.890 --> 00:59:52.470 align:middle line:84% People who are not affected are the bluish one. 00:59:52.470 --> 00:59:56.310 align:middle line:84% And you study a mechanism of transmission 00:59:56.310 --> 01:00:00.610 align:middle line:84% for a particular SNP, for a particular mutation. 01:00:00.610 --> 01:00:04.710 align:middle line:84% And what you do is use a quantity called [INAUDIBLE] 01:00:04.710 --> 01:00:06.960 align:middle line:90% ratio or loss cortex. 01:00:06.960 --> 01:00:08.710 align:middle line:90% What is this? 01:00:08.710 --> 01:00:12.270 align:middle line:84% You develop a method-- a model of transmission 01:00:12.270 --> 01:00:17.160 align:middle line:84% that is accusing that particular-- the SNP to be 01:00:17.160 --> 01:00:18.720 align:middle line:90% responsible for the disease. 01:00:18.720 --> 01:00:19.620 align:middle line:90% OK? 01:00:19.620 --> 01:00:23.610 align:middle line:84% So the tracking down of the SNP will follow-- 01:00:23.610 --> 01:00:25.890 align:middle line:84% the transmission of the SNP, of the mutation, 01:00:25.890 --> 01:00:28.830 align:middle line:84% will follow the pattern of the observable characters 01:00:28.830 --> 01:00:30.990 align:middle line:90% in the individuals you see. 01:00:30.990 --> 01:00:33.300 align:middle line:84% And then you make another one in which 01:00:33.300 --> 01:00:37.480 align:middle line:84% is-- in which the mutation is not responsible for it. 01:00:37.480 --> 01:00:41.280 align:middle line:84% And then using very simple statistical methods, 01:00:41.280 --> 01:00:44.490 align:middle line:84% you can compute that the data that you observe 01:00:44.490 --> 01:00:47.520 align:middle line:84% are generated by one model and the data 01:00:47.520 --> 01:00:50.130 align:middle line:84% that you observe are generated by the other model. 01:00:50.130 --> 01:00:52.650 align:middle line:84% And you can compare these two models, 01:00:52.650 --> 01:00:54.930 align:middle line:84% and you say, the model in which-- 01:00:54.930 --> 01:00:56.820 align:middle line:84% and you can say, the model in which 01:00:56.820 --> 01:01:01.530 align:middle line:84% this mutation is responsible for the disease, that 01:01:01.530 --> 01:01:04.230 align:middle line:84% is following the same pattern of inheritance, 01:01:04.230 --> 01:01:09.150 align:middle line:84% is n times more probable or less probable than the hypothesis 01:01:09.150 --> 01:01:13.880 align:middle line:84% in which the two patterns are not associated to each other. 01:01:13.880 --> 01:01:16.960 align:middle line:84% Now, the problem here is that, if you 01:01:16.960 --> 01:01:23.770 align:middle line:84% have a large pedigree of-- you have many SNPs, 01:01:23.770 --> 01:01:29.210 align:middle line:84% you end up in the problem of multiple comparisons. 01:01:29.210 --> 01:01:32.860 align:middle line:84% So I have to tell you this because this is my job, 01:01:32.860 --> 01:01:35.570 align:middle line:90% but I don't believe in this. 01:01:35.570 --> 01:01:38.560 align:middle line:84% But if you one day want to publish a paper using 01:01:38.560 --> 01:01:40.600 align:middle line:84% standard statistical, classical things, 01:01:40.600 --> 01:01:43.480 align:middle line:84% you will end up in a thing called-- 01:01:43.480 --> 01:01:47.690 align:middle line:84% in a muddy area called multiple comparisons. 01:01:47.690 --> 01:01:55.690 align:middle line:84% Now, multiple comparisons come from a classical superstition 01:01:55.690 --> 01:01:57.655 align:middle line:90% and from classical statistics. 01:01:57.655 --> 01:01:59.740 align:middle line:90% That is that you can infer-- 01:01:59.740 --> 01:02:03.280 align:middle line:84% you can say that two things are different if you can, 01:02:03.280 --> 01:02:05.680 align:middle line:84% by repeating the experiment a hundred times 01:02:05.680 --> 01:02:07.570 align:middle line:84% or a million times or some times, 01:02:07.570 --> 01:02:12.160 align:middle line:84% you will get less than 0.05 mistakes. 01:02:12.160 --> 01:02:13.990 align:middle line:90% You have heard of this p value. 01:02:13.990 --> 01:02:15.610 align:middle line:90% This is the p value. 01:02:15.610 --> 01:02:18.190 align:middle line:84% It is the probability that you will 01:02:18.190 --> 01:02:23.050 align:middle line:84% make an error by assuming that hypothesis more than n times. 01:02:23.050 --> 01:02:23.830 align:middle line:90% OK? 01:02:23.830 --> 01:02:26.780 align:middle line:84% In this case, it's five times out of a hundred. 01:02:26.780 --> 01:02:29.200 align:middle line:84% Now, if I'm testing one hypothesis, 01:02:29.200 --> 01:02:30.520 align:middle line:90% this is my error level. 01:02:30.520 --> 01:02:33.320 align:middle line:84% Supposedly, I am testing two hypotheses. 01:02:33.320 --> 01:02:37.750 align:middle line:84% Well, the probability that I will do this by chance 01:02:37.750 --> 01:02:40.720 align:middle line:84% doubles, so it's going to be the product of these two 01:02:40.720 --> 01:02:44.560 align:middle line:84% probabilities, right, because these are two joint probability 01:02:44.560 --> 01:02:45.400 align:middle line:90% distributions. 01:02:45.400 --> 01:02:51.610 align:middle line:84% Suppose I am observing 500, so it becomes 0.05 times 0.05-- 01:02:51.610 --> 01:02:52.480 align:middle line:90% smaller number. 01:02:52.480 --> 01:02:58.750 align:middle line:90% Suppose I have 500 SNPs. 01:02:58.750 --> 01:03:00.130 align:middle line:90% What should I do? 01:03:00.130 --> 01:03:03.640 align:middle line:84% That number will become so small that it would be impossible 01:03:03.640 --> 01:03:06.820 align:middle line:84% to prove anything, because the number would be-- 01:03:06.820 --> 01:03:09.670 align:middle line:84% because the probability that I pick up one hypothesis 01:03:09.670 --> 01:03:15.220 align:middle line:84% by chance would be so high that I will need a lot of evidence-- 01:03:15.220 --> 01:03:18.760 align:middle line:84% an unseemly amount of evidence to accept 01:03:18.760 --> 01:03:20.660 align:middle line:90% a particular hypothesis. 01:03:20.660 --> 01:03:24.070 align:middle line:84% So frequentist people have this problem. 01:03:24.070 --> 01:03:25.750 align:middle line:90% Patient people don't. 01:03:25.750 --> 01:03:27.520 align:middle line:90% But it's a problem. 01:03:27.520 --> 01:03:30.430 align:middle line:84% If you want to publish in a journal, 01:03:30.430 --> 01:03:32.470 align:middle line:84% this is a-- this is something you will-- 01:03:32.470 --> 01:03:40.080 align:middle line:90% 01:03:40.080 --> 01:03:41.220 align:middle line:90% OK. 01:03:41.220 --> 01:03:44.310 align:middle line:84% So this is nonparametric-- allele sharing 01:03:44.310 --> 01:03:47.130 align:middle line:84% is the thing I was telling you before, nonparametric method 01:03:47.130 --> 01:03:48.570 align:middle line:90% to assess linkage. 01:03:48.570 --> 01:03:52.410 align:middle line:84% And what you do is, you use siblings, 01:03:52.410 --> 01:03:55.470 align:middle line:84% and you assume that your particular SNP 01:03:55.470 --> 01:03:58.890 align:middle line:84% is identical-- the distribution of the SNP, of this mutation, 01:03:58.890 --> 01:04:00.840 align:middle line:90% is identical by descent. 01:04:00.840 --> 01:04:03.300 align:middle line:84% So I know what is the distribution I expect, 01:04:03.300 --> 01:04:07.290 align:middle line:84% and what I can do is to see, for a particular SNP, 01:04:07.290 --> 01:04:11.010 align:middle line:84% if there is a deviation from the distribution I would expect 01:04:11.010 --> 01:04:14.820 align:middle line:84% if these two guys were getting the way-- they were getting 01:04:14.820 --> 01:04:18.060 align:middle line:84% their mutation from the same pair of parents. 01:04:18.060 --> 01:04:18.600 align:middle line:90% Right? 01:04:18.600 --> 01:04:19.980 align:middle line:90% So these are siblings. 01:04:19.980 --> 01:04:25.090 align:middle line:84% I look at the probability that we'll get this SNP by chance. 01:04:25.090 --> 01:04:28.110 align:middle line:84% And if they don't get the SNP by chance because they're all 01:04:28.110 --> 01:04:30.900 align:middle line:84% affected, what we find out is that these people that 01:04:30.900 --> 01:04:32.760 align:middle line:84% are actually affected have the greater 01:04:32.760 --> 01:04:35.520 align:middle line:84% chance of getting this particular SNP, 01:04:35.520 --> 01:04:38.040 align:middle line:84% and this will make my SNP suspicious. 01:04:38.040 --> 01:04:41.850 align:middle line:84% Making it suspicious doesn't mean that you prove it. 01:04:41.850 --> 01:04:44.160 align:middle line:84% Now, if you want to prove something like this from 01:04:44.160 --> 01:04:47.160 align:middle line:84% a statistical point of view, using a non-parametric method-- 01:04:47.160 --> 01:04:49.860 align:middle line:84% because you make no assumption about distributional nature 01:04:49.860 --> 01:04:50.700 align:middle line:90% of your data-- 01:04:50.700 --> 01:04:54.850 align:middle line:84% you are going to have-- to need a huge amount of data. 01:04:54.850 --> 01:04:58.615 align:middle line:84% And this is not always feasible, but this 01:04:58.615 --> 01:04:59.740 align:middle line:90% is really, really feasible. 01:04:59.740 --> 01:05:03.540 align:middle line:84% So this is a weak test because it's simply telling you, 01:05:03.540 --> 01:05:06.660 align:middle line:84% the only thing you prove is that that particular allele is 01:05:06.660 --> 01:05:10.260 align:middle line:84% transmitted more than-- in a different way than random, 01:05:10.260 --> 01:05:13.650 align:middle line:84% well, more than its random population. 01:05:13.650 --> 01:05:16.110 align:middle line:84% Now, association studies are typically 01:05:16.110 --> 01:05:22.470 align:middle line:84% done using parametric methods, and they 01:05:22.470 --> 01:05:26.480 align:middle line:84% test for association between a particular phenotype 01:05:26.480 --> 01:05:32.610 align:middle line:84% and the genotype in two sets, in two different populations, 01:05:32.610 --> 01:05:34.440 align:middle line:90% well, in two different samples. 01:05:34.440 --> 01:05:35.520 align:middle line:90% What is the problem? 01:05:35.520 --> 01:05:37.380 align:middle line:84% It's that sometimes you recruit your sample 01:05:37.380 --> 01:05:40.090 align:middle line:90% from two different populations. 01:05:40.090 --> 01:05:41.860 align:middle line:84% I know it sounds absolutely silly, 01:05:41.860 --> 01:05:44.700 align:middle line:84% but this is one of the major concerns. 01:05:44.700 --> 01:05:48.030 align:middle line:84% Suppose the people in red are all people with asthma, 01:05:48.030 --> 01:05:51.420 align:middle line:84% and people with blue are all people without asthma. 01:05:51.420 --> 01:05:54.360 align:middle line:84% But then-- and they go and find out an association 01:05:54.360 --> 01:05:56.800 align:middle line:84% between the SNP and this-- and a mutation 01:05:56.800 --> 01:05:58.410 align:middle line:90% and then this phenotype. 01:05:58.410 --> 01:06:01.080 align:middle line:84% But then I discover that, for some reason, 01:06:01.080 --> 01:06:03.270 align:middle line:84% I have recruited everybody with asthma 01:06:03.270 --> 01:06:11.020 align:middle line:84% from Finland and everybody without asthma from Sicily. 01:06:11.020 --> 01:06:14.850 align:middle line:84% Now, is this SNP going to account for asthma, 01:06:14.850 --> 01:06:16.920 align:middle line:84% or is this SNP going to be account-- 01:06:16.920 --> 01:06:19.920 align:middle line:84% going to account for whatever this variance is 01:06:19.920 --> 01:06:21.565 align:middle line:90% between these two populations. 01:06:21.565 --> 01:06:23.190 align:middle line:84% These are called population admixtures, 01:06:23.190 --> 01:06:24.750 align:middle line:84% and I know it sounds silly for people 01:06:24.750 --> 01:06:30.660 align:middle line:84% to recruit things across different countries 01:06:30.660 --> 01:06:33.570 align:middle line:84% or buried, segregated populations. 01:06:33.570 --> 01:06:36.270 align:middle line:84% On the other hand, you may have-- 01:06:36.270 --> 01:06:39.630 align:middle line:84% you no guarantee that you don't have stratifications 01:06:39.630 --> 01:06:40.530 align:middle line:90% in your population. 01:06:40.530 --> 01:06:43.170 align:middle line:84% Even if you recruit in beautifully multicultural 01:06:43.170 --> 01:06:45.390 align:middle line:84% Massachusetts, the probability that 01:06:45.390 --> 01:06:49.470 align:middle line:84% you will find out something about the-- 01:06:49.470 --> 01:06:52.970 align:middle line:84% something about your sample that you don't really like, 01:06:52.970 --> 01:06:56.590 align:middle line:84% like a stratification, is not negligible. 01:06:56.590 --> 01:07:06.030 align:middle line:84% So what is really, really trendy today 01:07:06.030 --> 01:07:08.615 align:middle line:84% is to use transmission disequilibrium tests. 01:07:08.615 --> 01:07:09.990 align:middle line:84% Transmission disequilibrium tests 01:07:09.990 --> 01:07:11.990 align:middle line:84% are based on the assumption that you're actually 01:07:11.990 --> 01:07:15.325 align:middle line:84% using the parents of an individual as his controls, 01:07:15.325 --> 01:07:16.950 align:middle line:84% so you're sure you are not going to get 01:07:16.950 --> 01:07:19.020 align:middle line:90% any stratification, right? 01:07:19.020 --> 01:07:23.280 align:middle line:84% So I'm going to recruit a bunch of triads, father, mother, 01:07:23.280 --> 01:07:26.130 align:middle line:84% and child, and the child is affected 01:07:26.130 --> 01:07:30.270 align:middle line:84% and the parents are not affected, OK? 01:07:30.270 --> 01:07:33.000 align:middle line:84% And then what I'm going to do is to compute 01:07:33.000 --> 01:07:35.460 align:middle line:84% for each particular SNP, for a particular SNP, 01:07:35.460 --> 01:07:41.460 align:middle line:84% for a particular mutation, if there is a dependency, if there 01:07:41.460 --> 01:07:46.710 align:middle line:84% is a pattern, so to speak, between my distribution 01:07:46.710 --> 01:07:49.890 align:middle line:84% of transmission and the fact that all these kids are 01:07:49.890 --> 01:07:51.960 align:middle line:90% actually affected. 01:07:51.960 --> 01:07:53.830 align:middle line:90% And in this way. 01:07:53.830 --> 01:07:59.430 align:middle line:84% I have a very powerful test and non-certified controls. 01:07:59.430 --> 01:08:02.340 align:middle line:84% The only problem is that it's not always easy, 01:08:02.340 --> 01:08:05.070 align:middle line:84% as I was saying, to find parents for this. 01:08:05.070 --> 01:08:07.620 align:middle line:84% And the other problem is that you 01:08:07.620 --> 01:08:11.160 align:middle line:84% may find phenotypic stratification with that. 01:08:11.160 --> 01:08:15.450 align:middle line:84% Suppose you're looking for a particular disease that 01:08:15.450 --> 01:08:19.180 align:middle line:90% is related to weakness. 01:08:19.180 --> 01:08:22.810 align:middle line:90% An example I have is cholera. 01:08:22.810 --> 01:08:25.590 align:middle line:84% There are-- people believe that there is a SNP and people are 01:08:25.590 --> 01:08:29.970 align:middle line:90% looking for a SNP that is more-- 01:08:29.970 --> 01:08:32.130 align:middle line:84% making people more susceptible to have 01:08:32.130 --> 01:08:34.859 align:middle line:90% a bad outcome from cholera. 01:08:34.859 --> 01:08:37.109 align:middle line:84% And what they do is to go and recruit 01:08:37.109 --> 01:08:41.029 align:middle line:84% the people in households, so when somebody shows up 01:08:41.029 --> 01:08:43.640 align:middle line:84% in the hospital, they recruit the entire household 01:08:43.640 --> 01:08:45.470 align:middle line:90% and use the parents as control. 01:08:45.470 --> 01:08:48.380 align:middle line:84% Now, at that point, the problem is, 01:08:48.380 --> 01:08:51.377 align:middle line:84% because the phenotype is not really that easy to identify, 01:08:51.377 --> 01:08:53.210 align:middle line:84% we don't know if these people, for instance, 01:08:53.210 --> 01:08:57.229 align:middle line:84% got an immunity 20 years before because they got cholera. 01:08:57.229 --> 01:09:01.189 align:middle line:84% These are parents of a child who has-- who is 15 years old. 01:09:01.189 --> 01:09:04.250 align:middle line:84% In places where they have serious cholera, 01:09:04.250 --> 01:09:05.930 align:middle line:84% usually these are still young people, 01:09:05.930 --> 01:09:08.840 align:middle line:84% but from the societal structure, they're kind of old people. 01:09:08.840 --> 01:09:11.420 align:middle line:84% They have been there around for, like, 35 years. 01:09:11.420 --> 01:09:14.270 align:middle line:84% And this means that, at that point, 01:09:14.270 --> 01:09:17.720 align:middle line:84% you have this particular population that by design has 01:09:17.720 --> 01:09:19.970 align:middle line:84% a very different-- a phenotype that is 01:09:19.970 --> 01:09:23.520 align:middle line:90% very difficult to characterize. 01:09:23.520 --> 01:09:30.800 align:middle line:84% So QTL-- traits that, like in this case, 01:09:30.800 --> 01:09:32.630 align:middle line:90% variability in intensity. 01:09:32.630 --> 01:09:35.580 align:middle line:84% I'm not looking for something that is black and white. 01:09:35.580 --> 01:09:41.660 align:middle line:84% I'm looking for something with different degrees of severity. 01:09:41.660 --> 01:09:46.960 align:middle line:84% And until last year, there was really 01:09:46.960 --> 01:09:48.490 align:middle line:90% no way of doing these things. 01:09:48.490 --> 01:09:51.069 align:middle line:84% Well, nobody has proven ways of doing this. 01:09:51.069 --> 01:09:53.920 align:middle line:84% Now, people are working very hard 01:09:53.920 --> 01:09:57.040 align:middle line:84% because QTL are actually one of the most interesting things. 01:09:57.040 --> 01:10:00.700 align:middle line:84% People are looking really hard to find ways to characterize 01:10:00.700 --> 01:10:03.520 align:middle line:90% at least some type of QTL. 01:10:03.520 --> 01:10:09.560 align:middle line:84% And you can-- and complex QTL, like with census data. 01:10:09.560 --> 01:10:11.800 align:middle line:84% So one of the studies we have been doing 01:10:11.800 --> 01:10:16.180 align:middle line:90% is about the development of-- 01:10:16.180 --> 01:10:17.890 align:middle line:90% breast development in women. 01:10:17.890 --> 01:10:18.730 align:middle line:90% You say "why?" 01:10:18.730 --> 01:10:21.370 align:middle line:84% Well, because breast development has 01:10:21.370 --> 01:10:24.380 align:middle line:84% been associated to breast cancer, 01:10:24.380 --> 01:10:27.130 align:middle line:84% so we know that late breast development is 01:10:27.130 --> 01:10:29.800 align:middle line:84% protective of breast cancer, at least one 01:10:29.800 --> 01:10:31.700 align:middle line:90% type of breast cancer. 01:10:31.700 --> 01:10:35.960 align:middle line:90% Now this is a QTL. 01:10:35.960 --> 01:10:36.460 align:middle line:90% Why? 01:10:36.460 --> 01:10:42.610 align:middle line:84% Because you're looking at age, which is not a binary variable. 01:10:42.610 --> 01:10:48.010 align:middle line:84% And it's a very complicated QTL because it is-- 01:10:48.010 --> 01:10:49.870 align:middle line:90% you may have sensor data. 01:10:49.870 --> 01:10:52.690 align:middle line:84% You may have girls that at some point 01:10:52.690 --> 01:10:56.950 align:middle line:84% decide for one reason or another to drop out of the study. 01:10:56.950 --> 01:10:59.530 align:middle line:84% So the way which you can organize 01:10:59.530 --> 01:11:02.140 align:middle line:84% for this particular QTL-- not for every QTL-- is 01:11:02.140 --> 01:11:04.960 align:middle line:84% to imagine, to represent the structure as a survival 01:11:04.960 --> 01:11:09.340 align:middle line:84% study, in which you imagine that your SNP, a particular SNP, 01:11:09.340 --> 01:11:12.430 align:middle line:84% is going to be a treatment and the other 01:11:12.430 --> 01:11:13.870 align:middle line:90% is going to be the control. 01:11:13.870 --> 01:11:16.370 align:middle line:84% You draw your Kaplan-Meier curves, in this case, 01:11:16.370 --> 01:11:18.550 align:middle line:84% not for survival but for breast development, which 01:11:18.550 --> 01:11:20.570 align:middle line:90% is a kind of a deadline point. 01:11:20.570 --> 01:11:23.710 align:middle line:84% And at the end, you can compute the difference 01:11:23.710 --> 01:11:26.680 align:middle line:84% between these two samples and get evidence 01:11:26.680 --> 01:11:28.420 align:middle line:84% that actually there is a factor there 01:11:28.420 --> 01:11:29.810 align:middle line:90% that is putting them together. 01:11:29.810 --> 01:11:32.620 align:middle line:84% And you can use other multivariate structures 01:11:32.620 --> 01:11:40.480 align:middle line:84% that allow you to study the interaction of SNPs in QTL. 01:11:40.480 --> 01:11:44.500 align:middle line:84% But typically, if you want to have a general model for QTL, 01:11:44.500 --> 01:11:50.920 align:middle line:84% you have to resort to animals, animal models. 01:11:50.920 --> 01:11:53.200 align:middle line:84% Now there is another interesting thing 01:11:53.200 --> 01:11:56.910 align:middle line:84% that is happening on this street mostly, 01:11:56.910 --> 01:12:00.220 align:middle line:84% and has been happening for the past four or five years. 01:12:00.220 --> 01:12:09.270 align:middle line:84% There is-- we have two or three large phenotypic studies run 01:12:09.270 --> 01:12:12.540 align:middle line:84% by Harvard and affiliated hospitals. 01:12:12.540 --> 01:12:17.090 align:middle line:84% The oldest one is known as the Framingham Heart Study. 01:12:17.090 --> 01:12:20.145 align:middle line:84% Framingham Heart Study collects, I don't know, 50,000 people. 01:12:20.145 --> 01:12:23.280 align:middle line:84% They have been followed, and now they are the third generation. 01:12:23.280 --> 01:12:24.540 align:middle line:90% They have been followed. 01:12:24.540 --> 01:12:27.330 align:middle line:84% Their family members have been followed. 01:12:27.330 --> 01:12:32.455 align:middle line:84% We know about these people basically everything. 01:12:32.455 --> 01:12:36.160 align:middle line:84% They're-- these are selected for heart disease, 01:12:36.160 --> 01:12:37.410 align:middle line:90% but there are also studies-- 01:12:37.410 --> 01:12:38.993 align:middle line:84% there are cohort studies that have not 01:12:38.993 --> 01:12:41.370 align:middle line:84% been selected for anything, like the Nurses' Study. 01:12:41.370 --> 01:12:43.260 align:middle line:84% Nurses are wonderful individuals, 01:12:43.260 --> 01:12:45.940 align:middle line:84% incredibly compliant to doctor's order. 01:12:45.940 --> 01:12:49.440 align:middle line:84% So even if you harass them for 30 years, 01:12:49.440 --> 01:12:54.660 align:middle line:84% asking them every other year a complete report of what they 01:12:54.660 --> 01:12:58.500 align:middle line:84% have done, diseases they have, what is their diet, 01:12:58.500 --> 01:13:01.240 align:middle line:84% how much they weigh, they will comply. 01:13:01.240 --> 01:13:05.610 align:middle line:84% Nurses' Study has 150,000 women that, every other year, 01:13:05.610 --> 01:13:09.370 align:middle line:84% go and fill a questionnaire just for the sake of it. 01:13:09.370 --> 01:13:11.400 align:middle line:84% So about these women-- and this is now 01:13:11.400 --> 01:13:16.170 align:middle line:84% in the second generation-- we know absolutely everything. 01:13:16.170 --> 01:13:20.760 align:middle line:84% We know how much carrots they eat every year. 01:13:20.760 --> 01:13:24.390 align:middle line:84% And you probably see one-- the reason why 01:13:24.390 --> 01:13:27.270 align:middle line:84% you see these articles in the New York Times, 01:13:27.270 --> 01:13:31.200 align:middle line:84% they come out of chance, or the School of Public Health, 01:13:31.200 --> 01:13:33.120 align:middle line:84% but usually they mean chance, in this case. 01:13:33.120 --> 01:13:34.680 align:middle line:84% There are people who own this data. 01:13:34.680 --> 01:13:36.930 align:middle line:84% And these articles are like, you know, 01:13:36.930 --> 01:13:40.080 align:middle line:84% red hair are going to make you live longer. 01:13:40.080 --> 01:13:44.040 align:middle line:84% Or, you know, eating five carrots a day 01:13:44.040 --> 01:13:49.020 align:middle line:84% has been associated to having brighter eyes. 01:13:49.020 --> 01:13:52.566 align:middle line:84% Where those findings come out from? 01:13:52.566 --> 01:13:54.300 align:middle line:90% From the nurses study. 01:13:54.300 --> 01:13:56.880 align:middle line:84% There are these people that mine the data set. 01:13:56.880 --> 01:13:59.010 align:middle line:84% Again, they don't read in this case. 01:13:59.010 --> 01:14:00.850 align:middle line:84% They take a shower in the morning 01:14:00.850 --> 01:14:04.800 align:middle line:84% and say, what I can go and look into the nurses study? 01:14:04.800 --> 01:14:08.400 align:middle line:84% And they say, well, how about the association between mascara 01:14:08.400 --> 01:14:11.160 align:middle line:90% and blindness? 01:14:11.160 --> 01:14:14.430 align:middle line:84% We have 150,000 women that use mascara every day. 01:14:14.430 --> 01:14:16.830 align:middle line:84% Well, actually, you probably have a good proportion 01:14:16.830 --> 01:14:18.690 align:middle line:84% that uses mascara and a good proportion that 01:14:18.690 --> 01:14:20.017 align:middle line:90% doesn't use the mascara. 01:14:20.017 --> 01:14:21.000 align:middle line:90% And you know it. 01:14:21.000 --> 01:14:22.050 align:middle line:90% So they go there. 01:14:22.050 --> 01:14:24.148 align:middle line:84% And the joke on the street is that they don't 01:14:24.148 --> 01:14:25.440 align:middle line:90% publish in New England Journal. 01:14:25.440 --> 01:14:27.630 align:middle line:84% They publish directory on The New York Times 01:14:27.630 --> 01:14:31.110 align:middle line:84% because these are usually very high impact questions 01:14:31.110 --> 01:14:33.480 align:middle line:90% that people get right away. 01:14:33.480 --> 01:14:37.860 align:middle line:84% But imagine if you can take all those things, 01:14:37.860 --> 01:14:41.370 align:middle line:84% all those phenotypes, and you could genotype those women-- 01:14:41.370 --> 01:14:45.450 align:middle line:84% we have blood for about 90,000 of them-- 01:14:45.450 --> 01:14:52.320 align:middle line:84% and run full genetic scan for these women, what could you do? 01:14:52.320 --> 01:14:55.920 align:middle line:84% You could find an association between mascara and SNPs, 01:14:55.920 --> 01:15:00.180 align:middle line:84% or blindness and SNPs, or whatever and SNPs. 01:15:00.180 --> 01:15:02.320 align:middle line:90% Can we do this? 01:15:02.320 --> 01:15:04.890 align:middle line:90% Why don't we do it? 01:15:04.890 --> 01:15:11.310 align:middle line:84% Well, OK, genotype is net cost about $0.45, 01:15:11.310 --> 01:15:14.190 align:middle line:84% just to let you know where you are. 01:15:14.190 --> 01:15:20.940 align:middle line:84% To snip an individual map is going to cost-- so sorry, 01:15:20.940 --> 01:15:28.740 align:middle line:84% an individual map is going to be about 90,000, OK? 01:15:28.740 --> 01:15:34.470 align:middle line:84% So 1,000 individuals is going to cost you 90 million. 01:15:34.470 --> 01:15:38.340 align:middle line:90% 01:15:38.340 --> 01:15:41.790 align:middle line:84% And that money, they're very difficult to find 01:15:41.790 --> 01:15:45.120 align:middle line:90% in that structure. 01:15:45.120 --> 01:15:47.520 align:middle line:84% If you go to NIH and ask for this kind of money, 01:15:47.520 --> 01:15:50.370 align:middle line:90% you are going to get them. 01:15:50.370 --> 01:15:52.470 align:middle line:90% Work is not the problem. 01:15:52.470 --> 01:15:54.660 align:middle line:90% Postdocs at Harvard are cheap. 01:15:54.660 --> 01:15:56.010 align:middle line:90% And they work very hard. 01:15:56.010 --> 01:16:00.930 align:middle line:84% So we could actually have 1,000 of complete SNP 01:16:00.930 --> 01:16:04.260 align:middle line:84% maps in about 7,000 days, now, which 01:16:04.260 --> 01:16:07.920 align:middle line:84% is about two years for 10 postdocs, nothing. 01:16:07.920 --> 01:16:12.160 align:middle line:90% 01:16:12.160 --> 01:16:14.300 align:middle line:90% How can we solve this problem? 01:16:14.300 --> 01:16:18.890 align:middle line:84% Well, you remember when we said that these things go together? 01:16:18.890 --> 01:16:22.280 align:middle line:84% Maybe we don't need to genotype all of them. 01:16:22.280 --> 01:16:23.800 align:middle line:90% This is an example of a gene. 01:16:23.800 --> 01:16:27.670 align:middle line:90% I remember we-- 01:16:27.670 --> 01:16:29.860 align:middle line:90% I don't remember what gene. 01:16:29.860 --> 01:16:34.690 align:middle line:84% But this is a map of a gene that tells you the distance in terms 01:16:34.690 --> 01:16:37.300 align:middle line:84% of r-squared, which is the measure I was telling you 01:16:37.300 --> 01:16:40.010 align:middle line:90% before, the distance between-- 01:16:40.010 --> 01:16:43.670 align:middle line:84% the pairwise distance between two points. 01:16:43.670 --> 01:16:47.440 align:middle line:84% And as you can see, you can identify blocks there, 01:16:47.440 --> 01:16:52.480 align:middle line:84% the blue areas in which the correlation among the r-squared 01:16:52.480 --> 01:16:55.510 align:middle line:84% in this area is so high that you don't really 01:16:55.510 --> 01:16:59.290 align:middle line:90% need to genotype all of them. 01:16:59.290 --> 01:17:02.650 align:middle line:84% You can just decide where you go. 01:17:02.650 --> 01:17:04.390 align:middle line:84% And you say, OK, pick any of them. 01:17:04.390 --> 01:17:09.160 align:middle line:84% Problem is to know this, you first have to genotype them. 01:17:09.160 --> 01:17:12.070 align:middle line:84% So at that point, there is not much purpose for it. 01:17:12.070 --> 01:17:16.790 align:middle line:90% 01:17:16.790 --> 01:17:19.360 align:middle line:84% And the other problem is once you have this-- 01:17:19.360 --> 01:17:22.360 align:middle line:90% so here it's easy, right? 01:17:22.360 --> 01:17:23.590 align:middle line:90% When I have a region-- 01:17:23.590 --> 01:17:25.210 align:middle line:84% that region of recombination, there 01:17:25.210 --> 01:17:28.160 align:middle line:84% are four or five of lower recombination. 01:17:28.160 --> 01:17:30.730 align:middle line:84% I can say, OK, I pick up one, two, three, four. 01:17:30.730 --> 01:17:34.270 align:middle line:84% Now, the question is, which one I pick? 01:17:34.270 --> 01:17:36.040 align:middle line:84% Now, suppose they don't have five. 01:17:36.040 --> 01:17:40.930 align:middle line:84% I have in this case, 59, which are not that many. 01:17:40.930 --> 01:17:46.070 align:middle line:84% Regions like this there are 229 long. 01:17:46.070 --> 01:17:49.130 align:middle line:90% So well, your guess. 01:17:49.130 --> 01:17:51.347 align:middle line:84% How many SNPs do I need to genotype? 01:17:51.347 --> 01:17:52.430 align:middle line:90% Not which ones, of course. 01:17:52.430 --> 01:17:53.750 align:middle line:90% It would require a little. 01:17:53.750 --> 01:17:56.720 align:middle line:84% But how many do I need to genotype here 01:17:56.720 --> 01:17:58.590 align:middle line:84% to account for all the variations? 01:17:58.590 --> 01:18:00.440 align:middle line:90% So these are the SNPs. 01:18:00.440 --> 01:18:04.310 align:middle line:84% These are all the combination I get in the population, OK? 01:18:04.310 --> 01:18:05.550 align:middle line:90% These SNPs go together. 01:18:05.550 --> 01:18:08.450 align:middle line:84% So I don't expect to see all possible combinations. 01:18:08.450 --> 01:18:12.260 align:middle line:84% How many of them I will need to account 01:18:12.260 --> 01:18:16.880 align:middle line:84% for the 14 different alternative variations of [INAUDIBLE] 01:18:16.880 --> 01:18:20.960 align:middle line:90% over these 59 SNPs? 01:18:20.960 --> 01:18:23.390 align:middle line:90% I want the number. 01:18:23.390 --> 01:18:25.244 align:middle line:84% Jose, you're up to your third pizza. 01:18:25.244 --> 01:18:30.410 align:middle line:90% 01:18:30.410 --> 01:18:31.120 align:middle line:90% 15? 01:18:31.120 --> 01:18:31.830 align:middle line:90% AUDIENCE: 10. 01:18:31.830 --> 01:18:32.540 align:middle line:90% MARCO RAMONI: 10? 01:18:32.540 --> 01:18:34.370 align:middle line:90% OK, we have a 10 here. 01:18:34.370 --> 01:18:35.330 align:middle line:90% Anybody getting closer? 01:18:35.330 --> 01:18:40.680 align:middle line:90% 01:18:40.680 --> 01:18:41.610 align:middle line:90% That's a good guess. 01:18:41.610 --> 01:18:45.430 align:middle line:90% 01:18:45.430 --> 01:18:46.180 align:middle line:90% You will buy 10? 01:18:46.180 --> 01:18:48.950 align:middle line:90% 01:18:48.950 --> 01:18:49.550 align:middle line:90% Buy 10? 01:18:49.550 --> 01:18:53.570 align:middle line:90% 01:18:53.570 --> 01:18:54.070 align:middle line:90% Five. 01:18:54.070 --> 01:18:57.860 align:middle line:90% 01:18:57.860 --> 01:19:03.080 align:middle line:84% These five SNPs allow you to reconstruct all the variations 01:19:03.080 --> 01:19:05.720 align:middle line:90% without any information loss. 01:19:05.720 --> 01:19:08.130 align:middle line:84% I have a guy, Channing, a good friend of mine. 01:19:08.130 --> 01:19:11.570 align:middle line:84% And when we develop the method to identify this thing, 01:19:11.570 --> 01:19:15.770 align:middle line:84% bought me dinner because he has spent two years staring 01:19:15.770 --> 01:19:18.048 align:middle line:84% at these pictures and trying to figure out which were 01:19:18.048 --> 01:19:19.340 align:middle line:90% the SNPs that came [INAUDIBLE]. 01:19:19.340 --> 01:19:22.340 align:middle line:84% And actually, when he was wrong, getting a lot of shit 01:19:22.340 --> 01:19:24.540 align:middle line:84% from people because, oh, you got me the wrong thing. 01:19:24.540 --> 01:19:25.998 align:middle line:84% Now I mentioned it's a mind numbing 01:19:25.998 --> 01:19:30.080 align:middle line:84% task to stare at this picture and try to find out 01:19:30.080 --> 01:19:33.780 align:middle line:84% which are the optimal one to genotype. 01:19:33.780 --> 01:19:36.290 align:middle line:84% So what they have been doing is to play with the structure 01:19:36.290 --> 01:19:38.000 align:middle line:90% called haplotypes. 01:19:38.000 --> 01:19:40.095 align:middle line:84% Haplotypes are the things you get from your mother 01:19:40.095 --> 01:19:40.970 align:middle line:90% and from your father. 01:19:40.970 --> 01:19:42.387 align:middle line:84% And there is stretch of the genome 01:19:42.387 --> 01:19:44.660 align:middle line:84% that is consistent one chromosome on another. 01:19:44.660 --> 01:19:46.250 align:middle line:84% Now, the problem with haplotypes is 01:19:46.250 --> 01:19:49.550 align:middle line:84% that they're very difficult to identify for us. 01:19:49.550 --> 01:19:51.740 align:middle line:84% When you sequence somebody, you are 01:19:51.740 --> 01:19:55.370 align:middle line:84% going to have maps that look like the one I showed before. 01:19:55.370 --> 01:19:58.490 align:middle line:84% I can tell you if you're heterozygous, homozygous, 01:19:58.490 --> 01:20:02.690 align:middle line:84% or majority all homozygous, and of the minority of. 01:20:02.690 --> 01:20:06.120 align:middle line:90% But I cannot give you two bases. 01:20:06.120 --> 01:20:10.070 align:middle line:84% So if that particular SNP is A and T, 01:20:10.070 --> 01:20:16.850 align:middle line:84% as two alternative bases, I can tell if you are A-T, T-T, A-A. 01:20:16.850 --> 01:20:20.120 align:middle line:84% But I cannot tell you if you are T-A. 01:20:20.120 --> 01:20:25.520 align:middle line:84% I cannot decide if two things are coming both from 01:20:25.520 --> 01:20:27.680 align:middle line:84% your mother or both from your father, 01:20:27.680 --> 01:20:31.250 align:middle line:84% which is exactly the kind of information we need to build 01:20:31.250 --> 01:20:33.690 align:middle line:84% these kind of maps to reduce this information. 01:20:33.690 --> 01:20:36.500 align:middle line:84% So these structures are called haplotypes. 01:20:36.500 --> 01:20:39.890 align:middle line:90% And there are molecular methods. 01:20:39.890 --> 01:20:42.650 align:middle line:84% They are horribly expensive to do this. 01:20:42.650 --> 01:20:46.040 align:middle line:90% So they go into the chromosomes. 01:20:46.040 --> 01:20:48.590 align:middle line:84% And they tear apart the chromosome 01:20:48.590 --> 01:20:50.820 align:middle line:84% and return to separate chromosomes. 01:20:50.820 --> 01:20:52.820 align:middle line:84% And you know that one is coming from your mother 01:20:52.820 --> 01:20:55.250 align:middle line:84% and one is coming from your father. 01:20:55.250 --> 01:20:56.900 align:middle line:84% There are stochastic methods to do 01:20:56.900 --> 01:20:59.270 align:middle line:84% this in which you can try to figure out 01:20:59.270 --> 01:21:02.930 align:middle line:84% by making some assumption how these things are transmitted 01:21:02.930 --> 01:21:04.070 align:middle line:90% on a population. 01:21:04.070 --> 01:21:09.860 align:middle line:84% What is the distribution of an entire haplotype? 01:21:09.860 --> 01:21:13.040 align:middle line:84% But the best way of doing this is 01:21:13.040 --> 01:21:15.950 align:middle line:84% to use trios and try to guess what 01:21:15.950 --> 01:21:18.290 align:middle line:84% is the SNP that is coming from the mother 01:21:18.290 --> 01:21:20.690 align:middle line:84% and which is the one coming from the father. 01:21:20.690 --> 01:21:23.450 align:middle line:84% And so what people are doing now is 01:21:23.450 --> 01:21:28.490 align:middle line:84% developing a thing called HapMap in which we will have a map-- 01:21:28.490 --> 01:21:30.200 align:middle line:90% you can go to hapmap.org. 01:21:30.200 --> 01:21:32.630 align:middle line:84% And you will find all the data that's currently available. 01:21:32.630 --> 01:21:34.580 align:middle line:90% 01:21:34.580 --> 01:21:35.420 align:middle line:90% It's systematic. 01:21:35.420 --> 01:21:39.770 align:middle line:84% Now the resolution is 30,000 bases, 30 kilobases, 01:21:39.770 --> 01:21:45.230 align:middle line:84% a systematic genotyping of all SNPs on a relatively small 01:21:45.230 --> 01:21:45.830 align:middle line:90% population. 01:21:45.830 --> 01:21:48.530 align:middle line:84% There are 33 years in which we can actually 01:21:48.530 --> 01:21:51.663 align:middle line:90% find out which are the phases-- 01:21:51.663 --> 01:21:54.080 align:middle line:84% the fact that one is coming from the father and the mother 01:21:54.080 --> 01:21:56.330 align:middle line:84% is called the phase of a genotype. 01:21:56.330 --> 01:22:00.380 align:middle line:84% We can identify the phases from the parents. 01:22:00.380 --> 01:22:01.590 align:middle line:90% And why you do this? 01:22:01.590 --> 01:22:05.870 align:middle line:84% Well, because next time we want to design a genetic study, 01:22:05.870 --> 01:22:08.270 align:middle line:84% I don't have to run the genetic study 01:22:08.270 --> 01:22:11.360 align:middle line:84% and then say, oh, I could just genotype this one. 01:22:11.360 --> 01:22:13.520 align:middle line:84% I can actually go there, see I am 01:22:13.520 --> 01:22:15.200 align:middle line:84% interested in this particular gene, 01:22:15.200 --> 01:22:18.110 align:middle line:84% find out which are the SNPs that I really need to genotype. 01:22:18.110 --> 01:22:20.960 align:middle line:90% 01:22:20.960 --> 01:22:23.390 align:middle line:84% But there is some even more good news 01:22:23.390 --> 01:22:28.910 align:middle line:84% that apparently, all these things actually go together 01:22:28.910 --> 01:22:30.560 align:middle line:90% in stretches. 01:22:30.560 --> 01:22:32.540 align:middle line:84% So this is a paper from a couple of years 01:22:32.540 --> 01:22:38.240 align:middle line:84% ago in which they show that they made a high resolution 01:22:38.240 --> 01:22:42.080 align:middle line:84% map of a region of chromosome four associated 01:22:42.080 --> 01:22:43.740 align:middle line:90% with Crohn's disease. 01:22:43.740 --> 01:22:46.850 align:middle line:84% And what they found aside from the clinical things, what 01:22:46.850 --> 01:22:49.370 align:middle line:84% they found out is that this region is actually 01:22:49.370 --> 01:22:53.990 align:middle line:84% from an evolutionary point of view broken in 11 subregions. 01:22:53.990 --> 01:22:56.030 align:middle line:90% These regions are stable. 01:22:56.030 --> 01:22:58.220 align:middle line:84% They are transmitted together, so to speak. 01:22:58.220 --> 01:23:00.620 align:middle line:84% And they are interrupted once in a while 01:23:00.620 --> 01:23:03.260 align:middle line:84% by [INAUDIBLE] regional recombination. 01:23:03.260 --> 01:23:06.410 align:middle line:84% And the [INAUDIBLE] is they found out that if you look 01:23:06.410 --> 01:23:10.940 align:middle line:84% at all these variations of the alternative common haplotypes 01:23:10.940 --> 01:23:13.760 align:middle line:84% you have in the population, they actually come-- 01:23:13.760 --> 01:23:15.290 align:middle line:90% you see the different colors? 01:23:15.290 --> 01:23:20.390 align:middle line:84% They actually come from four ancestral haplotypes 01:23:20.390 --> 01:23:21.770 align:middle line:90% that are recombined. 01:23:21.770 --> 01:23:24.180 align:middle line:84% They've been recombined over the generations. 01:23:24.180 --> 01:23:26.040 align:middle line:84% So again, this gets back to your question. 01:23:26.040 --> 01:23:28.860 align:middle line:84% I'm not sure that there is, from a physical point of view, 01:23:28.860 --> 01:23:31.700 align:middle line:84% there is no-- you see the big block there 01:23:31.700 --> 01:23:34.190 align:middle line:90% in the middle, 92 kilobases? 01:23:34.190 --> 01:23:36.380 align:middle line:84% Well, I'm not entirely sure that there is not 01:23:36.380 --> 01:23:39.140 align:middle line:84% a very high recombination spot there. 01:23:39.140 --> 01:23:42.200 align:middle line:84% But because they all come from four haplotypes, 01:23:42.200 --> 01:23:45.230 align:middle line:84% with not other variation, I cannot see any recombination 01:23:45.230 --> 01:23:51.200 align:middle line:84% there because the probability of recombining requires a little 01:23:51.200 --> 01:23:52.410 align:middle line:90% pool to fish from. 01:23:52.410 --> 01:23:55.940 align:middle line:84% And in this case, they pool to fish from is very, very poor. 01:23:55.940 --> 01:23:58.200 align:middle line:84% Needless to say, these are all Caucasians. 01:23:58.200 --> 01:24:01.240 align:middle line:84% And they all come from this handful of individuals. 01:24:01.240 --> 01:24:06.960 align:middle line:84% So apparently, these people all come-- there's 129 SNP pairs-- 01:24:06.960 --> 01:24:08.460 align:middle line:90% not SNP pairs, trios. 01:24:08.460 --> 01:24:13.330 align:middle line:84% They all come from four ancestral haplotypes. 01:24:13.330 --> 01:24:15.690 align:middle line:90% The way you do is to use-- 01:24:15.690 --> 01:24:18.570 align:middle line:84% the way you identify this block is to use the Markov models. 01:24:18.570 --> 01:24:20.490 align:middle line:84% I will talk about the Markov models 01:24:20.490 --> 01:24:26.710 align:middle line:84% next week for the machine learning method. 01:24:26.710 --> 01:24:29.760 align:middle line:84% So I will not bother you with this. 01:24:29.760 --> 01:24:33.222 align:middle line:90% 01:24:33.222 --> 01:24:36.240 align:middle line:84% And at this point, what we can do once we have 01:24:36.240 --> 01:24:39.870 align:middle line:84% these blocks is to identify the SNPs that 01:24:39.870 --> 01:24:45.510 align:middle line:84% are tagging those particular blocks and then run them. 01:24:45.510 --> 01:24:49.890 align:middle line:84% And the same thing we've shown is an exponential saving 01:24:49.890 --> 01:24:51.100 align:middle line:90% with length. 01:24:51.100 --> 01:24:54.370 align:middle line:90% So it's about an average 10%. 01:24:54.370 --> 01:24:57.450 align:middle line:84% But if your block is very big, the number 01:24:57.450 --> 01:25:00.630 align:middle line:84% of SNPs you will need will be much smaller than 10%. 01:25:00.630 --> 01:25:05.730 align:middle line:84% And if it's a small block, the number of SNP you will need 01:25:05.730 --> 01:25:06.975 align:middle line:90% will be a bigger number. 01:25:06.975 --> 01:25:11.260 align:middle line:90% 01:25:11.260 --> 01:25:15.720 align:middle line:84% So the take home message, the fundamental take home message 01:25:15.720 --> 01:25:20.220 align:middle line:84% for a medical class is that all this technology 01:25:20.220 --> 01:25:25.840 align:middle line:84% has these beautiful properties of finding out stuff. 01:25:25.840 --> 01:25:28.740 align:middle line:84% But the only way to find out this stuff 01:25:28.740 --> 01:25:32.200 align:middle line:90% is to have good phenotypes. 01:25:32.200 --> 01:25:36.780 align:middle line:84% One day, running a SNP will cost $0.01. 01:25:36.780 --> 01:25:39.830 align:middle line:84% Maybe it will cost a tenth of a cent-- 01:25:39.830 --> 01:25:41.670 align:middle line:90% near nothing. 01:25:41.670 --> 01:25:44.340 align:middle line:84% But what makes very precious, what 01:25:44.340 --> 01:25:47.220 align:middle line:84% makes these studies possible, is the fact that you 01:25:47.220 --> 01:25:50.100 align:middle line:84% have something like the [INAUDIBLE] study 01:25:50.100 --> 01:25:53.730 align:middle line:84% that you have critically annotated, 01:25:53.730 --> 01:26:01.500 align:middle line:84% carefully recorded phenotypes and nicely characterized. 01:26:01.500 --> 01:26:03.960 align:middle line:84% And this is where the other side of medical knowledge 01:26:03.960 --> 01:26:07.110 align:middle line:84% comes in, not to find out what is really the cause, 01:26:07.110 --> 01:26:12.660 align:middle line:84% but just to identify precisely, as crisply as possible, 01:26:12.660 --> 01:26:14.880 align:middle line:90% a particular phenotype. 01:26:14.880 --> 01:26:17.850 align:middle line:84% At that point, you can actually map the phenotype. 01:26:17.850 --> 01:26:24.883 align:middle line:84% If your phenotype is slippery, yeah, well, it's 01:26:24.883 --> 01:26:26.550 align:middle line:84% very hard you will find any association. 01:26:26.550 --> 01:26:28.200 align:middle line:84% And if you find it, it probably will not 01:26:28.200 --> 01:26:29.575 align:middle line:84% be association with the thing you 01:26:29.575 --> 01:26:32.280 align:middle line:84% want because that's going to be a different phenotype, the one 01:26:32.280 --> 01:26:33.040 align:middle line:90% you're thinking. 01:26:33.040 --> 01:26:35.580 align:middle line:84% So the critical thing here is not, 01:26:35.580 --> 01:26:38.610 align:middle line:84% to go back to the very first slide, is not really 01:26:38.610 --> 01:26:40.470 align:middle line:90% to do bean bag genetics. 01:26:40.470 --> 01:26:43.440 align:middle line:84% The real critical thing here is to have good medicine that 01:26:43.440 --> 01:26:46.080 align:middle line:84% is able to characterize phenotypes that in turn allowed 01:26:46.080 --> 01:26:49.980 align:middle line:84% these studies to happen and to create good explanations 01:26:49.980 --> 01:26:51.750 align:middle line:90% for these phenotypes. 01:26:51.750 --> 01:26:54.450 align:middle line:84% And the other big take home message 01:26:54.450 --> 01:27:00.690 align:middle line:84% is that hypothesis-driven is out of fashion. 01:27:00.690 --> 01:27:03.060 align:middle line:84% We don't need hypothesis anymore. 01:27:03.060 --> 01:27:05.340 align:middle line:84% With enough money, we can have skills. 01:27:05.340 --> 01:27:06.450 align:middle line:90% We have the intellect. 01:27:06.450 --> 01:27:08.100 align:middle line:90% And we have phenotypes. 01:27:08.100 --> 01:27:10.110 align:middle line:90% We have enough information. 01:27:10.110 --> 01:27:14.340 align:middle line:84% We have enough juice to go and get our answer 01:27:14.340 --> 01:27:16.170 align:middle line:84% without thinking them in the shower, 01:27:16.170 --> 01:27:19.380 align:middle line:84% but looking systematically at the genome 01:27:19.380 --> 01:27:23.130 align:middle line:84% and taking advantage of the fact that something is more 01:27:23.130 --> 01:27:25.000 align:middle line:90% likely with something else. 01:27:25.000 --> 01:27:27.810 align:middle line:84% So if I tell you that something is associated with something 01:27:27.810 --> 01:27:31.380 align:middle line:84% else and this is my only study, we will need 10 years 01:27:31.380 --> 01:27:34.500 align:middle line:84% before people can consolidate the studies in a meta analysis 01:27:34.500 --> 01:27:37.140 align:middle line:84% study and say, oh, the evidence is not that big. 01:27:37.140 --> 01:27:40.530 align:middle line:84% But if I start telling you, I analyzed 01:27:40.530 --> 01:27:45.210 align:middle line:84% 500, or 5,000, or 50,000 SNPs and here it comes. 01:27:45.210 --> 01:27:48.780 align:middle line:84% This SNP is a million times more probable than any other SNP 01:27:48.780 --> 01:27:50.640 align:middle line:84% to be associated with this phenotype. 01:27:50.640 --> 01:27:54.210 align:middle line:84% That's a measure that you cannot get from an hypothesis-driven 01:27:54.210 --> 01:27:55.440 align:middle line:90% test. 01:27:55.440 --> 01:27:58.960 align:middle line:84% So OK, so when you will be reviewers of papers and grants, 01:27:58.960 --> 01:28:01.440 align:middle line:84% usually back people because their studies are not 01:28:01.440 --> 01:28:03.870 align:middle line:84% hypothesis-driven, remember this. 01:28:03.870 --> 01:28:05.720 align:middle line:90% See you next--