WEBVTT 00:00:00.000 --> 00:00:00.800 align:middle line:90% 00:00:00.800 --> 00:00:03.390 align:middle line:84% ISAAC SAMUEL KOHANE: So the overview of today's class-- 00:00:03.390 --> 00:00:07.050 align:middle line:84% I'm going to try to convince you that the future is now, 00:00:07.050 --> 00:00:10.680 align:middle line:84% that all this talk about the geniculate revolution 00:00:10.680 --> 00:00:13.290 align:middle line:84% is not pie in the sky stuff for venture 00:00:13.290 --> 00:00:15.520 align:middle line:84% capitalists for the next 10 years, 00:00:15.520 --> 00:00:19.440 align:middle line:84% but it's stuff that is of immediate practicality today. 00:00:19.440 --> 00:00:22.620 align:middle line:84% I'll spend a little bit of time in distinguishing 00:00:22.620 --> 00:00:25.770 align:middle line:84% between genomic and genetic more for 00:00:25.770 --> 00:00:29.760 align:middle line:84% your taxonomical edification and for any substantive reason. 00:00:29.760 --> 00:00:32.602 align:middle line:84% We have to touch upon heredity and what that is all about. 00:00:32.602 --> 00:00:34.560 align:middle line:84% And then we have to talk about the fundamentals 00:00:34.560 --> 00:00:38.370 align:middle line:84% of the resequencing of the diagnostic process that's 00:00:38.370 --> 00:00:42.100 align:middle line:84% being put into effect by the availability today of all 00:00:42.100 --> 00:00:44.100 align:middle line:84% this genomic data and then I'm going 00:00:44.100 --> 00:00:49.290 align:middle line:84% to finish up about how all this is resulting in accelerating 00:00:49.290 --> 00:00:50.370 align:middle line:90% consumer activation. 00:00:50.370 --> 00:00:53.810 align:middle line:84% And for those of you who are practitioners of medicine, 00:00:53.810 --> 00:00:55.710 align:middle line:84% and there's a subset of you who are, 00:00:55.710 --> 00:00:58.590 align:middle line:84% this should be alarming-- not that it's bad 00:00:58.590 --> 00:01:01.770 align:middle line:84% that consumers get involved but that they're ahead of us 00:01:01.770 --> 00:01:06.990 align:middle line:84% in their knowledge and application of this knowledge. 00:01:06.990 --> 00:01:08.580 align:middle line:90% So the future is now. 00:01:08.580 --> 00:01:12.640 align:middle line:84% So just because some of us, in fact, 00:01:12.640 --> 00:01:15.090 align:middle line:84% claim not to have taken any biology, 00:01:15.090 --> 00:01:17.490 align:middle line:84% I just want to review some basics. 00:01:17.490 --> 00:01:22.200 align:middle line:90% 00:01:22.200 --> 00:01:24.480 align:middle line:84% This will be old hat to some of you, 00:01:24.480 --> 00:01:28.980 align:middle line:84% but since you may not have that old hat, 00:01:28.980 --> 00:01:31.350 align:middle line:90% I'm going to review it. 00:01:31.350 --> 00:01:34.770 align:middle line:84% This is the old dogma, and this dogma is flawed. 00:01:34.770 --> 00:01:36.720 align:middle line:90% And it's not quite true. 00:01:36.720 --> 00:01:39.270 align:middle line:90% Hey, welcome. 00:01:39.270 --> 00:01:40.852 align:middle line:90% Just introduce yourself. 00:01:40.852 --> 00:01:41.943 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:01:41.943 --> 00:01:43.860 align:middle line:84% ISAAC SAMUEL KOHANE: And you are a student at? 00:01:43.860 --> 00:01:44.860 align:middle line:90% AUDIENCE: Oh, I'm sorry. 00:01:44.860 --> 00:01:46.240 align:middle line:90% I'm a student at MIT. 00:01:46.240 --> 00:01:47.070 align:middle line:90% HST student. 00:01:47.070 --> 00:01:49.710 align:middle line:84% ISAAC SAMUEL KOHANE: HST student, chemistry background. 00:01:49.710 --> 00:01:50.400 align:middle line:90% All right. 00:01:50.400 --> 00:01:53.490 align:middle line:84% Todd, have ever had a basic genetics course in your life? 00:01:53.490 --> 00:01:55.230 align:middle line:90% I ask the simple questions. 00:01:55.230 --> 00:01:55.890 align:middle line:90% Yes? 00:01:55.890 --> 00:01:56.550 align:middle line:90% AUDIENCE: Yeah. 00:01:56.550 --> 00:01:56.970 align:middle line:90% ISAAC SAMUEL KOHANE: All right. 00:01:56.970 --> 00:01:57.450 align:middle line:90% AUDIENCE: Undergrad. 00:01:57.450 --> 00:01:58.742 align:middle line:90% ISAAC SAMUEL KOHANE: Undergrad. 00:01:58.742 --> 00:02:00.420 align:middle line:90% All right. 00:02:00.420 --> 00:02:09.090 align:middle line:84% So I'm just reviewing the basic dogma, which is DNA gives rise 00:02:09.090 --> 00:02:13.350 align:middle line:84% to transcribed RNA, which gives rise to translated protein. 00:02:13.350 --> 00:02:15.450 align:middle line:84% So let me ask a question of the class, 00:02:15.450 --> 00:02:18.075 align:middle line:84% and I start saying you because I can't remember your names yet. 00:02:18.075 --> 00:02:21.240 align:middle line:90% 00:02:21.240 --> 00:02:23.880 align:middle line:84% Is the DNA in each cell in our body the same? 00:02:23.880 --> 00:02:26.670 align:middle line:90% 00:02:26.670 --> 00:02:27.540 align:middle line:90% AUDIENCE: No. 00:02:27.540 --> 00:02:29.630 align:middle line:84% ISAAC SAMUEL KOHANE: Why is it different? 00:02:29.630 --> 00:02:31.560 align:middle line:84% AUDIENCE: Which as the cells differentiate 00:02:31.560 --> 00:02:32.820 align:middle line:90% making it every time. 00:02:32.820 --> 00:02:34.200 align:middle line:84% ISAAC SAMUEL KOHANE: Good because you just stated 00:02:34.200 --> 00:02:36.240 align:middle line:84% the misunderstanding that most people have. 00:02:36.240 --> 00:02:43.020 align:middle line:84% In fact, with exception of your sperm or eggs, 00:02:43.020 --> 00:02:48.180 align:middle line:84% all the genome of the entire body is exactly the same. 00:02:48.180 --> 00:02:49.930 align:middle line:90% The full genome is available. 00:02:49.930 --> 00:02:52.470 align:middle line:84% In fact, the reason you can clone an animal 00:02:52.470 --> 00:02:55.650 align:middle line:84% from any somatic cell in the body in theory 00:02:55.650 --> 00:02:58.350 align:middle line:84% is because with some caveats which 00:02:58.350 --> 00:03:00.300 align:middle line:84% I don't want to get into right now, 00:03:00.300 --> 00:03:02.160 align:middle line:84% it's, in fact, the same darn genome. 00:03:02.160 --> 00:03:05.370 align:middle line:84% The exceptions are gametes, which have half of the genome, 00:03:05.370 --> 00:03:11.860 align:middle line:84% and red blood cells, which have no nucleus. 00:03:11.860 --> 00:03:13.413 align:middle line:90% Now why is that important? 00:03:13.413 --> 00:03:15.330 align:middle line:84% And there's some other things like methylation 00:03:15.330 --> 00:03:17.913 align:middle line:84% and the tips of chromosomes which I don't 00:03:17.913 --> 00:03:19.080 align:middle line:90% want to give you details of. 00:03:19.080 --> 00:03:20.170 align:middle line:90% But why is that important? 00:03:20.170 --> 00:03:21.628 align:middle line:84% Because in fact, what makes a liver 00:03:21.628 --> 00:03:26.070 align:middle line:84% cell different from a white blood cell is not the DNA 00:03:26.070 --> 00:03:29.940 align:middle line:84% but what RNA is being transcribed 00:03:29.940 --> 00:03:36.630 align:middle line:84% and also what protein is being translated from the RNA. 00:03:36.630 --> 00:03:39.300 align:middle line:84% And the reason I bring this up is 00:03:39.300 --> 00:03:43.020 align:middle line:84% because as we measure DNA using sequencing or genotyping 00:03:43.020 --> 00:03:46.890 align:middle line:84% or we're measuring RNAs using transcriptional profiling or we 00:03:46.890 --> 00:03:51.060 align:middle line:84% measure translation using proteomics, 00:03:51.060 --> 00:03:53.250 align:middle line:84% we have to start thinking now, even 00:03:53.250 --> 00:03:55.110 align:middle line:90% those of us who are not MDs. 00:03:55.110 --> 00:03:57.240 align:middle line:84% As clinicians, where can we actually obtain 00:03:57.240 --> 00:03:59.430 align:middle line:90% this biological material? 00:03:59.430 --> 00:04:02.510 align:middle line:84% And if you're doing a DNA study, anything is good. 00:04:02.510 --> 00:04:03.760 align:middle line:90% You can get white blood cells. 00:04:03.760 --> 00:04:04.960 align:middle line:90% You can get a blood sample. 00:04:04.960 --> 00:04:08.070 align:middle line:84% But if you wanted to study, for instance, brains 00:04:08.070 --> 00:04:11.130 align:middle line:84% and understand what makes someone susceptible to brain 00:04:11.130 --> 00:04:15.990 align:middle line:84% tumors, you could not get an RNA sample 00:04:15.990 --> 00:04:18.850 align:middle line:84% without actually getting a hunk of someone's brain. 00:04:18.850 --> 00:04:22.350 align:middle line:84% And so that immediately tells you some limitations 00:04:22.350 --> 00:04:23.410 align:middle line:90% of these technologies. 00:04:23.410 --> 00:04:31.080 align:middle line:84% So if you want to do RNA or protein profiling 00:04:31.080 --> 00:04:34.492 align:middle line:84% to look at someone's likelihood to get a disease, 00:04:34.492 --> 00:04:36.450 align:middle line:84% for instance, you have to get the right tissue, 00:04:36.450 --> 00:04:38.610 align:middle line:84% and that tells you the kind of limitations 00:04:38.610 --> 00:04:40.883 align:middle line:84% that we face when we're doing clinical studies. 00:04:40.883 --> 00:04:42.300 align:middle line:84% I mean, certainly, we can hack out 00:04:42.300 --> 00:04:44.400 align:middle line:84% brains out of mice or drosophila, 00:04:44.400 --> 00:04:48.420 align:middle line:84% but we can't readily do that with humans. 00:04:48.420 --> 00:04:52.620 align:middle line:84% And yet that's frustrating because on this end, 00:04:52.620 --> 00:04:55.320 align:middle line:84% we're closer to function than we are to this end. 00:04:55.320 --> 00:04:59.460 align:middle line:84% This is the master code, but this 00:04:59.460 --> 00:05:01.120 align:middle line:84% is what leads to the actual functioning 00:05:01.120 --> 00:05:03.180 align:middle line:90% the interacting of the proteins. 00:05:03.180 --> 00:05:06.490 align:middle line:84% And so people are very excited about proteomics, for instance, 00:05:06.490 --> 00:05:07.990 align:middle line:84% because it's very close to function. 00:05:07.990 --> 00:05:10.448 align:middle line:84% But when you're looking about what are we going to actually 00:05:10.448 --> 00:05:12.960 align:middle line:84% be measuring the proteins on, we're 00:05:12.960 --> 00:05:15.330 align:middle line:84% fairly limited to things like serum and urine. 00:05:15.330 --> 00:05:18.900 align:middle line:84% We can't take pieces of tissue out of human beings. 00:05:18.900 --> 00:05:22.650 align:middle line:84% And we can talk on another day about how 00:05:22.650 --> 00:05:25.450 align:middle line:90% to get around these obstacles. 00:05:25.450 --> 00:05:28.800 align:middle line:84% So what's the magnitude of the task? 00:05:28.800 --> 00:05:31.530 align:middle line:84% From one perspective, it seems relatively piddling. 00:05:31.530 --> 00:05:34.290 align:middle line:84% I mean, there's, after all, 46 chromosomes, 00:05:34.290 --> 00:05:37.650 align:middle line:84% and these chromosomes are these little microscopic things, 00:05:37.650 --> 00:05:40.410 align:middle line:84% bushy things of hyper coiled DNA. 00:05:40.410 --> 00:05:43.800 align:middle line:90% And it seems fairly discrete. 00:05:43.800 --> 00:05:46.230 align:middle line:84% And I know that you know it's 3 billion bases, 00:05:46.230 --> 00:05:48.660 align:middle line:84% but I want to give you an appreciation for how 00:05:48.660 --> 00:05:51.160 align:middle line:90% much 3 billion bases are. 00:05:51.160 --> 00:05:55.230 align:middle line:84% So let's say we have a necklace of beads. 00:05:55.230 --> 00:05:59.378 align:middle line:84% Each of those beads represents one base pair. 00:05:59.378 --> 00:06:01.920 align:middle line:84% Can I take it for granted that y'all know what the base pairs 00:06:01.920 --> 00:06:02.802 align:middle line:90% are? 00:06:02.802 --> 00:06:04.890 align:middle line:90% ACTG, right. 00:06:04.890 --> 00:06:05.490 align:middle line:90% Good. 00:06:05.490 --> 00:06:11.130 align:middle line:84% So if you took a bunch of these necklaces 00:06:11.130 --> 00:06:17.790 align:middle line:84% and gave a necklace to everyone in Shea Stadium, 00:06:17.790 --> 00:06:22.680 align:middle line:84% it will still take 1,000 stadia to get enough necklaces 00:06:22.680 --> 00:06:25.930 align:middle line:84% to get all the bases that we have in the human genome. 00:06:25.930 --> 00:06:29.010 align:middle line:84% So that's a lot of beads, and it gives you 00:06:29.010 --> 00:06:34.800 align:middle line:84% some sense of the coding challenge and the deciphering 00:06:34.800 --> 00:06:35.520 align:middle line:90% challenge. 00:06:35.520 --> 00:06:38.380 align:middle line:90% 00:06:38.380 --> 00:06:42.190 align:middle line:84% But you'll have technology nerds say the following. 00:06:42.190 --> 00:06:46.930 align:middle line:84% Some say that Madonna's music is actually more information 00:06:46.930 --> 00:06:49.342 align:middle line:84% than Madonna the person because-- 00:06:49.342 --> 00:06:51.550 align:middle line:84% and I actually heard this from-- someone in Microsoft 00:06:51.550 --> 00:06:56.590 align:middle line:84% say this in a big [INAUDIBLE] talk because they said, look. 00:06:56.590 --> 00:07:03.340 align:middle line:84% A DVD with her music video is 4.7 gigabytes 00:07:03.340 --> 00:07:08.740 align:middle line:84% whereas her genome, 3 billion base pairs 00:07:08.740 --> 00:07:10.390 align:middle line:84% at two bits per base, since there's 00:07:10.390 --> 00:07:16.400 align:middle line:84% four bases, and 8 bits per byte, is only 3/4 of a gigabyte. 00:07:16.400 --> 00:07:19.730 align:middle line:84% So what are we making the big deal about? 00:07:19.730 --> 00:07:22.180 align:middle line:84% And in fact, they went on to say correctly 00:07:22.180 --> 00:07:29.380 align:middle line:84% that every human is on average only less than 0.1% different 00:07:29.380 --> 00:07:30.430 align:middle line:90% than any other human. 00:07:30.430 --> 00:07:33.100 align:middle line:84% So we could actually store you all as just a difference 00:07:33.100 --> 00:07:35.358 align:middle line:90% file of everybody else. 00:07:35.358 --> 00:07:37.900 align:middle line:84% So you can it turns out if you did that, you can compress you 00:07:37.900 --> 00:07:41.860 align:middle line:84% down to the size that could fit nicely on an obsolete diskette. 00:07:41.860 --> 00:07:47.650 align:middle line:84% But that's a the question that I want to raise. 00:07:47.650 --> 00:07:51.890 align:middle line:90% Is Madonna her DNA sequence? 00:07:51.890 --> 00:07:53.770 align:middle line:84% And the answer, I hope you realize, 00:07:53.770 --> 00:07:57.550 align:middle line:84% is no because what makes Madonna particularly Madonna 00:07:57.550 --> 00:08:01.360 align:middle line:84% is not only the way she developed in her womb 00:08:01.360 --> 00:08:03.850 align:middle line:84% even without environmental influences as programmed 00:08:03.850 --> 00:08:06.610 align:middle line:84% but all the developmental influences she 00:08:06.610 --> 00:08:07.540 align:middle line:90% had through her life-- 00:08:07.540 --> 00:08:10.660 align:middle line:84% the good knocks, the bad knocks, all those experiences, 00:08:10.660 --> 00:08:14.590 align:middle line:84% environmental exposures that bring her state, 00:08:14.590 --> 00:08:18.950 align:middle line:84% her organismal state, into the current megastar state it is. 00:08:18.950 --> 00:08:21.640 align:middle line:84% And there are many other states that 00:08:21.640 --> 00:08:23.140 align:middle line:84% could have been depending on what 00:08:23.140 --> 00:08:25.390 align:middle line:84% those environmental influences were. 00:08:25.390 --> 00:08:28.780 align:middle line:84% And if you really want to capture a state, 00:08:28.780 --> 00:08:30.460 align:middle line:90% it's not only-- 00:08:30.460 --> 00:08:35.140 align:middle line:84% it's therefore the evolution of her physical state, 00:08:35.140 --> 00:08:38.289 align:middle line:84% not just her genome, but all the alternative splicing. 00:08:38.289 --> 00:08:39.880 align:middle line:84% As you'll hear about next lecture, 00:08:39.880 --> 00:08:42.880 align:middle line:84% genes can be spliced in multiple ways to create RNA. 00:08:42.880 --> 00:08:46.270 align:middle line:84% So that's on average about three per gene. 00:08:46.270 --> 00:08:49.210 align:middle line:84% The proteins which are derived from RNA 00:08:49.210 --> 00:08:55.120 align:middle line:84% can have a multitude of different modifications-- 00:08:55.120 --> 00:08:57.190 align:middle line:84% adding of glucose, glycosylation, 00:08:57.190 --> 00:08:59.175 align:middle line:84% adding of phosphorous, phosphorylation. 00:08:59.175 --> 00:09:01.300 align:middle line:84% And in fact, this is underestimated I'm giving you. 00:09:01.300 --> 00:09:05.170 align:middle line:84% That's on order of 100 to 1,000 modifications. 00:09:05.170 --> 00:09:07.630 align:middle line:84% So that's greatly increasing the complexity. 00:09:07.630 --> 00:09:10.990 align:middle line:84% And then in a multi-cellular organism, 00:09:10.990 --> 00:09:12.610 align:middle line:84% each of those gene products can be 00:09:12.610 --> 00:09:14.890 align:middle line:84% in any particular compartment in the cell 00:09:14.890 --> 00:09:20.410 align:middle line:84% and then can float around to any cell in the body 00:09:20.410 --> 00:09:21.620 align:middle line:90% on the order of trillions. 00:09:21.620 --> 00:09:24.850 align:middle line:84% So Madonna now is several orders of magnitude 00:09:24.850 --> 00:09:27.400 align:middle line:84% larger than her music, as we are all. 00:09:27.400 --> 00:09:30.760 align:middle line:84% And this is just as a caveat to note that although it's 00:09:30.760 --> 00:09:34.090 align:middle line:84% really impressive that we can sequence an entire genome, 00:09:34.090 --> 00:09:39.340 align:middle line:84% all we're showing is the master code that 00:09:39.340 --> 00:09:41.650 align:middle line:84% can result in one of many potential fates, 00:09:41.650 --> 00:09:45.130 align:middle line:84% and to avoid the reductionism that often happens, 00:09:45.130 --> 00:09:46.982 align:middle line:84% you have to remind yourself what were 00:09:46.982 --> 00:09:48.190 align:middle line:90% the environmental influences. 00:09:48.190 --> 00:09:51.310 align:middle line:84% And we'll come back to that again. 00:09:51.310 --> 00:09:55.980 align:middle line:84% Now getting on, again, a more optimistic technological note, 00:09:55.980 --> 00:10:01.630 align:middle line:84% it is nonetheless true that we can now measure today 00:10:01.630 --> 00:10:04.810 align:middle line:84% with commodity technology large parts 00:10:04.810 --> 00:10:08.560 align:middle line:84% of your genome for commodity prices. 00:10:08.560 --> 00:10:09.920 align:middle line:90% And what do I mean by that? 00:10:09.920 --> 00:10:12.713 align:middle line:84% Let's look at RNA microarrays, and I know some of about these 00:10:12.713 --> 00:10:13.630 align:middle line:90% and some of you don't. 00:10:13.630 --> 00:10:15.730 align:middle line:84% But I just want to remind you of them. 00:10:15.730 --> 00:10:16.870 align:middle line:90% If you take a tissue-- 00:10:16.870 --> 00:10:17.980 align:middle line:90% let's say, liver. 00:10:17.980 --> 00:10:20.260 align:middle line:84% Or tissue-- by the way, let me just 00:10:20.260 --> 00:10:25.420 align:middle line:84% interrupt myself to say that all these PowerPoint 00:10:25.420 --> 00:10:27.430 align:middle line:90% presentations-- thank you. 00:10:27.430 --> 00:10:30.250 align:middle line:84% All these PowerPoint presentations and the video 00:10:30.250 --> 00:10:33.520 align:middle line:84% are going to be available through the MyCourses website. 00:10:33.520 --> 00:10:36.760 align:middle line:84% If you don't have an account yet through MyCourses, 00:10:36.760 --> 00:10:38.508 align:middle line:90% you can definitely get one. 00:10:38.508 --> 00:10:40.300 align:middle line:84% And so all the PowerPoint and all the video 00:10:40.300 --> 00:10:41.240 align:middle line:90% will be available. 00:10:41.240 --> 00:10:46.330 align:middle line:84% So don't feel compelled to take everything down, and just 00:10:46.330 --> 00:10:49.390 align:middle line:84% be more sure that you're asking questions in the flesh. 00:10:49.390 --> 00:10:52.690 align:middle line:84% So take liver or, let's say, liver under influence 00:10:52.690 --> 00:10:54.250 align:middle line:90% of, perhaps, insulin. 00:10:54.250 --> 00:10:54.910 align:middle line:90% You take it. 00:10:54.910 --> 00:10:57.070 align:middle line:90% You grind it up. 00:10:57.070 --> 00:11:01.690 align:middle line:84% You label the RNA that was extracted 00:11:01.690 --> 00:11:06.157 align:middle line:84% from that ground up liver with a fluorescent compound. 00:11:06.157 --> 00:11:07.990 align:middle line:84% That means you attach a fluorescent compound 00:11:07.990 --> 00:11:11.110 align:middle line:90% to that labeled RNA. 00:11:11.110 --> 00:11:13.450 align:middle line:90% And then you purchase. 00:11:13.450 --> 00:11:15.395 align:middle line:84% And I used to say purchase or make, 00:11:15.395 --> 00:11:17.020 align:middle line:84% but that's becoming less and less true. 00:11:17.020 --> 00:11:19.030 align:middle line:84% It truly has become a commodity, and companies 00:11:19.030 --> 00:11:22.900 align:middle line:84% such as Affymetrix or Agilent will sell you these chips, 00:11:22.900 --> 00:11:25.330 align:middle line:84% and they can do it probably with higher quality, 00:11:25.330 --> 00:11:28.000 align:middle line:84% unfortunately, than we can in the homebrew 00:11:28.000 --> 00:11:30.070 align:middle line:84% fashion even though if you go to the Stanford 00:11:30.070 --> 00:11:32.320 align:middle line:84% website, microarray website, they'll 00:11:32.320 --> 00:11:34.930 align:middle line:84% tell you how to build your own microarray manufacturing 00:11:34.930 --> 00:11:38.500 align:middle line:84% facility out of RadioShack parts for less than $10,000. 00:11:38.500 --> 00:11:45.820 align:middle line:84% Nonetheless, it's easier and more standardized at this time 00:11:45.820 --> 00:11:50.440 align:middle line:84% to unfortunately buy it from this Microsoft-like monopoly 00:11:50.440 --> 00:11:52.180 align:middle line:84% called [? Affymetrix, ?] for instance. 00:11:52.180 --> 00:11:55.960 align:middle line:84% In any case, so you buy this chip, 00:11:55.960 --> 00:12:00.550 align:middle line:84% and pre-positioned on each spot on this chip-- 00:12:00.550 --> 00:12:03.250 align:middle line:84% and I will have a whole lecture devoted to microarrays, 00:12:03.250 --> 00:12:05.470 align:middle line:84% so don't worry about the details too much. 00:12:05.470 --> 00:12:07.630 align:middle line:84% Pre-positioned on spot on this microarray 00:12:07.630 --> 00:12:12.580 align:middle line:84% is a sequence of DNA corresponding to the gene 00:12:12.580 --> 00:12:13.700 align:middle line:90% that you're looking for. 00:12:13.700 --> 00:12:17.590 align:middle line:84% So this might be the sequence of DNA corresponding to growth 00:12:17.590 --> 00:12:22.420 align:middle line:84% hormone, a sequence of DNA corresponding to insulin, 00:12:22.420 --> 00:12:25.060 align:middle line:84% a sequence of DNA corresponding to the insulin receptor. 00:12:25.060 --> 00:12:29.890 align:middle line:84% And what's going to happen is through the hybridization 00:12:29.890 --> 00:12:34.330 align:middle line:84% reaction that you all learned in basic genetics, 00:12:34.330 --> 00:12:36.640 align:middle line:84% the RNA that you extract from the tissue 00:12:36.640 --> 00:12:40.090 align:middle line:84% will hybridize with the matching sequence of DNA 00:12:40.090 --> 00:12:43.300 align:middle line:90% that is present on this chip. 00:12:43.300 --> 00:12:45.310 align:middle line:84% And suddenly, the chips are no bigger than this 00:12:45.310 --> 00:12:46.510 align:middle line:90% if not smaller. 00:12:46.510 --> 00:12:50.470 align:middle line:84% And then you scan it using a simple HP scanner of no higher 00:12:50.470 --> 00:12:53.600 align:middle line:84% quality than one that you used to scan in your family photos. 00:12:53.600 --> 00:12:55.520 align:middle line:84% You scan the image of a chip, and it 00:12:55.520 --> 00:12:59.990 align:middle line:84% fluoresces proportional to the amount of fluorescent RNA 00:12:59.990 --> 00:13:01.100 align:middle line:90% at any given spot. 00:13:01.100 --> 00:13:02.900 align:middle line:84% And so what you have is a readout 00:13:02.900 --> 00:13:06.320 align:middle line:84% now at every spot of how much of that 00:13:06.320 --> 00:13:09.680 align:middle line:90% RNA was present in that sample. 00:13:09.680 --> 00:13:12.260 align:middle line:84% And the current density of Affymetrix microwaves 00:13:12.260 --> 00:13:14.300 align:middle line:90% is now on a single chip. 00:13:14.300 --> 00:13:17.630 align:middle line:84% They have spotted-- and this is using, actually, 00:13:17.630 --> 00:13:19.190 align:middle line:90% a photolithographic process. 00:13:19.190 --> 00:13:24.230 align:middle line:84% They have spotted every single gene in the human genome-- 00:13:24.230 --> 00:13:26.660 align:middle line:84% in fact, so many so that they are actually 00:13:26.660 --> 00:13:30.680 align:middle line:84% putting extra ones, such as different alternative splices. 00:13:30.680 --> 00:13:36.620 align:middle line:84% So more than 30,000 genes on one chip, and you 00:13:36.620 --> 00:13:40.880 align:middle line:84% can buy it through Harvard pricing for about $250. 00:13:40.880 --> 00:13:43.490 align:middle line:84% So you can measure the expression 00:13:43.490 --> 00:13:47.342 align:middle line:90% of 30,000 genes for $250. 00:13:47.342 --> 00:13:49.550 align:middle line:84% Now that sounds academically as an interesting thing, 00:13:49.550 --> 00:13:54.510 align:middle line:84% but how does that translate into genomic medicine? 00:13:54.510 --> 00:13:59.060 align:middle line:84% So let me give you my poster child for a pulse cycle study 00:13:59.060 --> 00:14:01.880 align:middle line:90% which has launched 1,000 chips-- 00:14:01.880 --> 00:14:04.340 align:middle line:90% chips, but chips as well. 00:14:04.340 --> 00:14:06.410 align:middle line:84% It's launched several million chips. 00:14:06.410 --> 00:14:09.590 align:middle line:90% 00:14:09.590 --> 00:14:11.755 align:middle line:84% This was a study that was done out of Stanford. 00:14:11.755 --> 00:14:13.130 align:middle line:84% And they did the following thing. 00:14:13.130 --> 00:14:15.880 align:middle line:84% They had patients with large B-cell lymphoma. 00:14:15.880 --> 00:14:18.230 align:middle line:84% Large B-cell lymphoma is a type of cancer 00:14:18.230 --> 00:14:21.320 align:middle line:84% that is on a sort of medium bad disease-- not as 00:14:21.320 --> 00:14:25.790 align:middle line:84% bad as lung disease, not as benign as thyroid disease, 00:14:25.790 --> 00:14:26.990 align:middle line:90% thyroid cancer. 00:14:26.990 --> 00:14:31.400 align:middle line:84% And this is a disease that was previously monolithic. 00:14:31.400 --> 00:14:33.800 align:middle line:84% In other words, the patient came to you, 00:14:33.800 --> 00:14:35.660 align:middle line:84% and you look on the microscope and said, 00:14:35.660 --> 00:14:37.460 align:middle line:84% ah, you have large B-cell lymphoma, 00:14:37.460 --> 00:14:39.793 align:middle line:84% and you had a rough estimate of how long they were going 00:14:39.793 --> 00:14:41.930 align:middle line:84% to live, and you'd give them all the same treatment 00:14:41.930 --> 00:14:44.210 align:middle line:84% because you could not distinguish between subclasses 00:14:44.210 --> 00:14:45.960 align:middle line:90% of these patients. 00:14:45.960 --> 00:14:48.350 align:middle line:84% So what they did through techniques 00:14:48.350 --> 00:14:50.270 align:middle line:84% that you will learn about in this class 00:14:50.270 --> 00:14:54.980 align:middle line:84% is they performed clustering analysis on these patients 00:14:54.980 --> 00:14:58.880 align:middle line:84% when they took RNA extracted out of the lymphoma 00:14:58.880 --> 00:15:01.970 align:middle line:90% of several dozen patients. 00:15:01.970 --> 00:15:03.830 align:middle line:84% And when they did a clustering experiment, 00:15:03.830 --> 00:15:08.000 align:middle line:84% they were able to see that there were two groups of patients 00:15:08.000 --> 00:15:10.940 align:middle line:84% based solely on their gene expression profile 00:15:10.940 --> 00:15:15.530 align:middle line:84% that there was one group of patients that have one gene 00:15:15.530 --> 00:15:17.990 align:middle line:84% expression profile and another group of patients that 00:15:17.990 --> 00:15:20.090 align:middle line:84% had a different gene expression profile. 00:15:20.090 --> 00:15:22.760 align:middle line:84% Then and only then, they asked themselves 00:15:22.760 --> 00:15:25.610 align:middle line:84% after the fact what was different about these patients. 00:15:25.610 --> 00:15:29.120 align:middle line:84% And very rewardingly, they found that corresponding 00:15:29.120 --> 00:15:31.310 align:middle line:84% to 1 gene expression profile-- in other words, one 00:15:31.310 --> 00:15:33.230 align:middle line:84% set of genes being switched on or off-- 00:15:33.230 --> 00:15:37.010 align:middle line:84% was this group of patients in red where here on the x-axis 00:15:37.010 --> 00:15:38.960 align:middle line:90% is years of survival. 00:15:38.960 --> 00:15:42.000 align:middle line:84% Here on the y-axis is probability of survival. 00:15:42.000 --> 00:15:44.390 align:middle line:84% And so what you see here is one group 00:15:44.390 --> 00:15:46.940 align:middle line:84% of patients who die very fast, most of them, 00:15:46.940 --> 00:15:48.980 align:middle line:84% by two years and one group of patients 00:15:48.980 --> 00:15:55.190 align:middle line:84% who have better than 50% survival in the 12 year 00:15:55.190 --> 00:15:55.970 align:middle line:90% outcome. 00:15:55.970 --> 00:15:58.220 align:middle line:84% And that was remarkable because it told us 00:15:58.220 --> 00:16:01.880 align:middle line:84% that previously, where we had not had a sub-diagnosis, 00:16:01.880 --> 00:16:04.130 align:middle line:84% we have now a new diagnosis-- high risk 00:16:04.130 --> 00:16:07.490 align:middle line:84% and low risk large B-cell lymphoma. 00:16:07.490 --> 00:16:11.150 align:middle line:90% Two, we have a new prognosis. 00:16:11.150 --> 00:16:13.340 align:middle line:84% We can tell patients, unfortunately, this profile, 00:16:13.340 --> 00:16:15.410 align:middle line:84% you're at much higher risk than you would 00:16:15.410 --> 00:16:16.520 align:middle line:90% have told them previously. 00:16:16.520 --> 00:16:20.040 align:middle line:84% Three, you have a new clinical opportunity, 00:16:20.040 --> 00:16:22.490 align:middle line:84% a new therapeutic opportunity, because without discovering 00:16:22.490 --> 00:16:27.560 align:middle line:84% any new drugs, you're just going to hit these ones much harder 00:16:27.560 --> 00:16:31.670 align:middle line:84% with the chemotherapeutic agents that you already have 00:16:31.670 --> 00:16:35.210 align:middle line:84% because you know they essentially have a much more 00:16:35.210 --> 00:16:36.860 align:middle line:90% dangerous kind of cancer. 00:16:36.860 --> 00:16:40.160 align:middle line:84% And then fourth, you have a new research opportunity. 00:16:40.160 --> 00:16:42.020 align:middle line:84% Why are these patients dying off? 00:16:42.020 --> 00:16:44.780 align:middle line:84% Are they pumping out-- do they have a transporter that's 00:16:44.780 --> 00:16:49.355 align:middle line:84% pumping out chemotherapeutic agents out of the cell? 00:16:49.355 --> 00:16:52.230 align:middle line:90% 00:16:52.230 --> 00:16:53.640 align:middle line:84% It's unclear, but we already know 00:16:53.640 --> 00:16:56.460 align:middle line:84% that they're using this as research to-- for instance, 00:16:56.460 --> 00:16:58.860 align:middle line:84% it turns out this is a B-cell marker. 00:16:58.860 --> 00:17:01.290 align:middle line:84% This group is enriched for B-cell markers. 00:17:01.290 --> 00:17:03.060 align:middle line:84% B-cells are a type of immune cell, 00:17:03.060 --> 00:17:04.890 align:middle line:84% and they're actually using a therapy 00:17:04.890 --> 00:17:08.500 align:middle line:84% directed against that B-cell to actually treat these patients. 00:17:08.500 --> 00:17:09.720 align:middle line:90% So that's fairly remarkable. 00:17:09.720 --> 00:17:12.750 align:middle line:84% From a trivial clinical research experiment using this commodity 00:17:12.750 --> 00:17:16.109 align:middle line:84% chip, we have a new diagnosis, a new prognosis, 00:17:16.109 --> 00:17:21.810 align:middle line:84% a new therapeutic opportunity, and a new research opportunity. 00:17:21.810 --> 00:17:26.099 align:middle line:84% And this has been reproduced multiple times for lung cancer, 00:17:26.099 --> 00:17:27.000 align:middle line:90% for breast cancer. 00:17:27.000 --> 00:17:30.122 align:middle line:84% You can now actually purchase a test in the Netherlands 00:17:30.122 --> 00:17:31.830 align:middle line:84% based on the needle biopsy of the breast. 00:17:31.830 --> 00:17:35.400 align:middle line:84% We'll actually stratify the woman on her prognosis 00:17:35.400 --> 00:17:36.870 align:middle line:90% based on the needle biopsy. 00:17:36.870 --> 00:17:39.760 align:middle line:90% And it's not only for cancers. 00:17:39.760 --> 00:17:42.780 align:middle line:84% Now it's been done for a variety of inflammatory conditions-- 00:17:42.780 --> 00:17:47.010 align:middle line:84% inflammatory bowel disease, rheumatoid arthritis, 00:17:47.010 --> 00:17:50.447 align:middle line:84% a whole bunch of conditions where you're actually 00:17:50.447 --> 00:17:52.530 align:middle line:84% getting a much broader insight into the physiology 00:17:52.530 --> 00:17:54.750 align:middle line:84% by looking at 30,000 genes than we were 00:17:54.750 --> 00:17:57.360 align:middle line:84% by just a patient asking a few questions 00:17:57.360 --> 00:18:01.420 align:middle line:84% and measuring just one or two variables in their blood. 00:18:01.420 --> 00:18:04.050 align:middle line:84% So this brings us to this cartoon 00:18:04.050 --> 00:18:06.420 align:middle line:84% that says you're not ill yet, Mr. [? Blondel, ?] 00:18:06.420 --> 00:18:08.200 align:middle line:90% but you've got potential. 00:18:08.200 --> 00:18:11.408 align:middle line:84% And of course, this is the motivation 00:18:11.408 --> 00:18:13.200 align:middle line:84% behind a lot of genomic medicine that we're 00:18:13.200 --> 00:18:16.170 align:middle line:84% going to be able to by having these highly 00:18:16.170 --> 00:18:19.890 align:middle line:84% forecasting markers, these predictive markers, 00:18:19.890 --> 00:18:22.890 align:middle line:84% we'll be able to move away from acute and expensive and not 00:18:22.890 --> 00:18:26.970 align:middle line:84% very effective interventions to much more cost 00:18:26.970 --> 00:18:29.910 align:middle line:84% effective, cheaper, more effective 00:18:29.910 --> 00:18:34.380 align:middle line:90% prevention techniques. 00:18:34.380 --> 00:18:36.750 align:middle line:90% And it's not theoretical. 00:18:36.750 --> 00:18:40.820 align:middle line:84% Right now one of the largest groups-- 00:18:40.820 --> 00:18:42.570 align:middle line:84% by the way, how does chemotherapy actually 00:18:42.570 --> 00:18:43.545 align:middle line:90% get devised? 00:18:43.545 --> 00:18:45.420 align:middle line:84% So how do these [INAUDIBLE] get devised? 00:18:45.420 --> 00:18:48.540 align:middle line:84% So there are these large cooperative oncology groups, 00:18:48.540 --> 00:18:52.050 align:middle line:84% literally dozens of hospitals and medical centers 00:18:52.050 --> 00:18:55.110 align:middle line:84% that cooperate on devising how much of cisplatin, 00:18:55.110 --> 00:18:57.570 align:middle line:84% how much Adriamycin, how much azathioprine 00:18:57.570 --> 00:19:00.180 align:middle line:84% to give to various patients with the various chemotherapy 00:19:00.180 --> 00:19:00.800 align:middle line:90% protocols? 00:19:00.800 --> 00:19:03.300 align:middle line:84% And they treat these thousands of patients, see how they do, 00:19:03.300 --> 00:19:06.870 align:middle line:84% and then issue a new protocol based on outcome. 00:19:06.870 --> 00:19:08.910 align:middle line:84% Very highly funded on the order of hundreds 00:19:08.910 --> 00:19:10.350 align:middle line:90% of millions of dollars a year. 00:19:10.350 --> 00:19:13.440 align:middle line:84% And today, because I'm part now of the cancer and leukemia 00:19:13.440 --> 00:19:19.080 align:middle line:84% group B, one of these cooperative groups is now-- 00:19:19.080 --> 00:19:21.270 align:middle line:84% this entire group is now coming up 00:19:21.270 --> 00:19:24.150 align:middle line:84% with new protocols that are all based 00:19:24.150 --> 00:19:27.613 align:middle line:84% on genome-wide understanding of expression. 00:19:27.613 --> 00:19:29.280 align:middle line:84% In other words, they're going to put you 00:19:29.280 --> 00:19:32.400 align:middle line:84% into a different treatment category based on your gene 00:19:32.400 --> 00:19:34.230 align:middle line:90% expression profile. 00:19:34.230 --> 00:19:36.660 align:middle line:84% And this should worry you because as you have learned 00:19:36.660 --> 00:19:40.682 align:middle line:84% in this class, there's a lot about this analysis, 00:19:40.682 --> 00:19:41.640 align:middle line:90% this cluster analysis-- 00:19:41.640 --> 00:19:44.590 align:middle line:84% I gave you the shining great story about it. 00:19:44.590 --> 00:19:46.600 align:middle line:84% But even that initial analysis I showed you 00:19:46.600 --> 00:19:48.420 align:middle line:84% of large B-cell lymphoma is flawed. 00:19:48.420 --> 00:19:51.510 align:middle line:84% And yet people are moving ahead very fast 00:19:51.510 --> 00:19:55.230 align:middle line:84% to turn these into clinical use diagnostics. 00:19:55.230 --> 00:19:57.540 align:middle line:84% And because they don't understand fundamentally 00:19:57.540 --> 00:19:59.100 align:middle line:84% the limitations of their clustering techniques, 00:19:59.100 --> 00:20:01.140 align:middle line:84% for instance, and the limitations of the measurement 00:20:01.140 --> 00:20:02.700 align:middle line:84% platforms like the microarray analysis, 00:20:02.700 --> 00:20:04.242 align:middle line:84% they're actually going to do patients 00:20:04.242 --> 00:20:06.840 align:middle line:84% a disservice by taking patients who don't need the therapy 00:20:06.840 --> 00:20:10.270 align:middle line:84% and put them into a high risk class and vise versa. 00:20:10.270 --> 00:20:11.680 align:middle line:84% So I think it's hugely important. 00:20:11.680 --> 00:20:14.040 align:middle line:84% One of the things that I want you to learn in this class 00:20:14.040 --> 00:20:18.610 align:middle line:84% is how to read skeptically the literature because in fact, 00:20:18.610 --> 00:20:20.640 align:middle line:84% it will help you become better leaders 00:20:20.640 --> 00:20:23.622 align:middle line:84% in this area of high dimensional medicine. 00:20:23.622 --> 00:20:25.080 align:middle line:84% What do I mean by high dimensional? 00:20:25.080 --> 00:20:26.460 align:middle line:84% Well, you're looking at thousands and thousands 00:20:26.460 --> 00:20:27.180 align:middle line:90% of variables. 00:20:27.180 --> 00:20:28.680 align:middle line:84% Well, here, we're just looking at expression. 00:20:28.680 --> 00:20:30.347 align:middle line:84% But as we'll get to later in the course, 00:20:30.347 --> 00:20:33.750 align:middle line:84% it's, of course, looking at all the different SNPs 00:20:33.750 --> 00:20:35.130 align:middle line:90% on your genome-- 00:20:35.130 --> 00:20:37.500 align:middle line:84% the different variations, I should say-- 00:20:37.500 --> 00:20:40.155 align:middle line:90% and your protein state. 00:20:40.155 --> 00:20:43.090 align:middle line:90% 00:20:43.090 --> 00:20:45.742 align:middle line:84% So this is the current state of the art 00:20:45.742 --> 00:20:46.950 align:middle line:90% for those of you not doctors. 00:20:46.950 --> 00:20:50.160 align:middle line:84% It's the stage of the cancer, where 00:20:50.160 --> 00:20:52.750 align:middle line:84% is it in your body, the grade, how ugly do the cells look 00:20:52.750 --> 00:20:54.750 align:middle line:84% under the microscope, and what kind of cell type 00:20:54.750 --> 00:20:58.860 align:middle line:84% are they is typically what drives diagnosis today. 00:20:58.860 --> 00:21:00.960 align:middle line:84% But even today, it's now starting 00:21:00.960 --> 00:21:02.970 align:middle line:84% to be driven in real protocols being 00:21:02.970 --> 00:21:07.260 align:middle line:84% deployed for real patients by expression profiling. 00:21:07.260 --> 00:21:10.050 align:middle line:84% And that begs the question that I 00:21:10.050 --> 00:21:14.310 align:middle line:84% think is now upon us, which is, are we 00:21:14.310 --> 00:21:18.420 align:middle line:84% now on the threshold of a new taxonomy of human diseases? 00:21:18.420 --> 00:21:22.290 align:middle line:84% Human disease has been classified according 00:21:22.290 --> 00:21:28.470 align:middle line:84% to manifestation, and it's actually not so long 00:21:28.470 --> 00:21:34.080 align:middle line:84% ago that we used to think of fever as a disease rather 00:21:34.080 --> 00:21:36.250 align:middle line:90% than a symptom. 00:21:36.250 --> 00:21:39.840 align:middle line:84% And if you look very closely at most diseases 00:21:39.840 --> 00:21:44.080 align:middle line:84% like the inflammatory diseases like arthritis, 00:21:44.080 --> 00:21:45.580 align:middle line:84% they're still, for the most part, 00:21:45.580 --> 00:21:51.600 align:middle line:84% based on phenomenology and not on some deep mechanistic 00:21:51.600 --> 00:21:54.630 align:middle line:90% etiology or cause. 00:21:54.630 --> 00:21:57.240 align:middle line:84% And as Thomas Lewis pointed out in 1944, 00:21:57.240 --> 00:21:58.800 align:middle line:84% the diagnosis of most human disease 00:21:58.800 --> 00:22:02.430 align:middle line:84% provides only insecure and temporary conceptions because 00:22:02.430 --> 00:22:04.230 align:middle line:84% in fact, what we're describing is always 00:22:04.230 --> 00:22:06.030 align:middle line:84% the tip of the iceberg, the things 00:22:06.030 --> 00:22:07.450 align:middle line:84% that the patients complain about. 00:22:07.450 --> 00:22:09.030 align:middle line:84% And there's only so many things you can complain about. 00:22:09.030 --> 00:22:10.488 align:middle line:84% This hurts, this hurts, this hurts, 00:22:10.488 --> 00:22:13.230 align:middle line:90% and I can or cannot do this. 00:22:13.230 --> 00:22:16.920 align:middle line:84% And there's a limited set of all the ways it can manifest. 00:22:16.920 --> 00:22:20.610 align:middle line:84% But it's actually reflecting a much larger multitude 00:22:20.610 --> 00:22:22.500 align:middle line:84% of possible things that can go wrong 00:22:22.500 --> 00:22:26.713 align:middle line:84% in this very complex mechanism that is the human being. 00:22:26.713 --> 00:22:28.380 align:middle line:84% And in fact, of the main common diseases 00:22:28.380 --> 00:22:30.720 align:middle line:84% only infectious diseases have a truly mechanism-based 00:22:30.720 --> 00:22:31.680 align:middle line:90% nomenclature. 00:22:31.680 --> 00:22:35.370 align:middle line:84% You see how people name infectious diseases. 00:22:35.370 --> 00:22:37.780 align:middle line:84% It goes right down to the organism causing it. 00:22:37.780 --> 00:22:40.620 align:middle line:84% The rest of it is pretty loosey goosey 00:22:40.620 --> 00:22:42.100 align:middle line:90% as it relates to the mechanism. 00:22:42.100 --> 00:22:45.240 align:middle line:84% And here we have part of the challenge in genomic medicine 00:22:45.240 --> 00:22:47.790 align:middle line:84% is if we're measuring comprehensively 00:22:47.790 --> 00:22:52.480 align:middle line:84% the entirety of the physiology of the patient state, 00:22:52.480 --> 00:22:55.050 align:middle line:84% we can actually come with a much more objective 00:22:55.050 --> 00:22:59.280 align:middle line:84% and universal nomenclature, which 00:22:59.280 --> 00:23:01.965 align:middle line:84% is a nontrivial thing for medical students 00:23:01.965 --> 00:23:03.150 align:middle line:90% who are being-- 00:23:03.150 --> 00:23:04.200 align:middle line:90% for those of you-- 00:23:04.200 --> 00:23:05.783 align:middle line:84% is there any medical students in here? 00:23:05.783 --> 00:23:10.025 align:middle line:84% Yes, who are being assaulted by the-- 00:23:10.025 --> 00:23:11.400 align:middle line:84% I was told once in medical school 00:23:11.400 --> 00:23:14.100 align:middle line:84% that I had to learn on the order of 20,000 new terms, 00:23:14.100 --> 00:23:19.300 align:middle line:84% and you can see they're already outdated. 00:23:19.300 --> 00:23:24.270 align:middle line:84% There's also a reason why today we 00:23:24.270 --> 00:23:26.220 align:middle line:84% should be thinking about changing 00:23:26.220 --> 00:23:28.530 align:middle line:90% the use of everyday medications. 00:23:28.530 --> 00:23:35.010 align:middle line:84% Shown here is a protein that codes for a potassium channel 00:23:35.010 --> 00:23:37.920 align:middle line:84% it's a protein that allows potassium 00:23:37.920 --> 00:23:41.880 align:middle line:84% in or out in a selective fashion in and out of cells. 00:23:41.880 --> 00:23:47.220 align:middle line:84% And it turns out if you have a misspelling or a polymorphism 00:23:47.220 --> 00:23:50.070 align:middle line:84% of one of your nucleotides here, then 00:23:50.070 --> 00:23:53.310 align:middle line:84% it will result in a new amino acid in one spot, 00:23:53.310 --> 00:23:57.540 align:middle line:84% and that will change the electrical properties 00:23:57.540 --> 00:24:04.110 align:middle line:84% of this potassium channel such that individuals are much more 00:24:04.110 --> 00:24:09.240 align:middle line:84% prone to cardiac arrhythmias, bad rhythms of their heart, 00:24:09.240 --> 00:24:13.050 align:middle line:90% if they are given a sulfa drug. 00:24:13.050 --> 00:24:15.270 align:middle line:84% Now sulfa drugs, for those that you don't know, 00:24:15.270 --> 00:24:17.340 align:middle line:84% are a broad range of antibiotics, 00:24:17.340 --> 00:24:20.620 align:middle line:84% and in pediatrics, for instance, we give almost all kids sulfa 00:24:20.620 --> 00:24:21.120 align:middle line:90% drugs. 00:24:21.120 --> 00:24:24.720 align:middle line:84% Bactrim has part of it a sulfa drug. 00:24:24.720 --> 00:24:26.730 align:middle line:84% And so what we're saying here is there 00:24:26.730 --> 00:24:32.140 align:middle line:84% is a subset of children who have this mutation. 00:24:32.140 --> 00:24:36.060 align:middle line:84% And if you give them sulfa drugs today, you're 00:24:36.060 --> 00:24:38.250 align:middle line:84% going to put them at risk of death, 00:24:38.250 --> 00:24:42.690 align:middle line:84% which seems like a bad decision for just treatment of earache. 00:24:42.690 --> 00:24:45.030 align:middle line:84% Now if you're still awake, you should be asking yourself 00:24:45.030 --> 00:24:45.690 align:middle line:90% the following. 00:24:45.690 --> 00:24:49.173 align:middle line:84% Well, Zack, you're telling me about this, 00:24:49.173 --> 00:24:51.090 align:middle line:84% but why should I really be worried about this? 00:24:51.090 --> 00:24:52.830 align:middle line:90% It's obviously a rare thing. 00:24:52.830 --> 00:24:55.650 align:middle line:84% It's probably less than 1 in 10,000. 00:24:55.650 --> 00:24:57.990 align:middle line:84% And why should I do this expensive re-engineering 00:24:57.990 --> 00:25:04.680 align:middle line:84% of the medical process in order to be able to avoid 00:25:04.680 --> 00:25:06.460 align:middle line:90% harming these children? 00:25:06.460 --> 00:25:07.960 align:middle line:84% So I'm going to ask you a question. 00:25:07.960 --> 00:25:10.500 align:middle line:84% How much do you think in a high throughput laboratory 00:25:10.500 --> 00:25:13.560 align:middle line:84% like the Channing Laboratory across the street, 00:25:13.560 --> 00:25:20.100 align:middle line:84% how much do you think it costs to see if your genome has 00:25:20.100 --> 00:25:21.930 align:middle line:84% this polymorphism, and therefore, we 00:25:21.930 --> 00:25:24.180 align:middle line:84% should be careful about treating you with sulfa drugs? 00:25:24.180 --> 00:25:28.230 align:middle line:84% How much do you think it would cost us to process your blood 00:25:28.230 --> 00:25:34.710 align:middle line:84% and obtain which spelling you had for that one polymorphism? 00:25:34.710 --> 00:25:36.870 align:middle line:84% And I've asked this of the entire class, 00:25:36.870 --> 00:25:39.870 align:middle line:84% and I've asked this endless times for very senior people 00:25:39.870 --> 00:25:41.800 align:middle line:90% and very junior people. 00:25:41.800 --> 00:25:44.560 align:middle line:84% And unless you really know the answer, then don't answer. 00:25:44.560 --> 00:25:48.280 align:middle line:90% But let's go for it. 00:25:48.280 --> 00:25:49.555 align:middle line:90% "Price is Right" rules work. 00:25:49.555 --> 00:25:52.180 align:middle line:84% AUDIENCE: Well, you assume that you have a blood sample already 00:25:52.180 --> 00:25:52.660 align:middle line:90% there? 00:25:52.660 --> 00:25:54.010 align:middle line:84% Just the cost of running the test? 00:25:54.010 --> 00:25:54.820 align:middle line:84% ISAAC SAMUEL KOHANE: Yes, I'm not asking 00:25:54.820 --> 00:25:56.590 align:middle line:90% you to pay for the blood draw. 00:25:56.590 --> 00:25:57.575 align:middle line:90% Right. 00:25:57.575 --> 00:25:58.075 align:middle line:90% How much? 00:25:58.075 --> 00:26:00.960 align:middle line:90% 00:26:00.960 --> 00:26:04.610 align:middle line:90% AUDIENCE: 30 bucks [INAUDIBLE]. 00:26:04.610 --> 00:26:06.710 align:middle line:84% ISAAC SAMUEL KOHANE: Less than a buck. 00:26:06.710 --> 00:26:07.880 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:26:07.880 --> 00:26:08.540 align:middle line:90% ISAAC SAMUEL KOHANE: 130 bucks. 00:26:08.540 --> 00:26:09.530 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:26:09.530 --> 00:26:11.978 align:middle line:90% ISAAC SAMUEL KOHANE: 5 bucks? 00:26:11.978 --> 00:26:13.400 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:26:13.400 --> 00:26:13.840 align:middle line:90% ISAAC SAMUEL KOHANE: How much? 00:26:13.840 --> 00:26:14.220 align:middle line:90% AUDIENCE: $1 00:26:14.220 --> 00:26:15.312 align:middle line:90% ISAAC SAMUEL KOHANE: $1. 00:26:15.312 --> 00:26:16.023 align:middle line:90% AUDIENCE: 50? 00:26:16.023 --> 00:26:17.106 align:middle line:90% ISAAC SAMUEL KOHANE: What? 00:26:17.106 --> 00:26:18.504 align:middle line:90% 50? 00:26:18.504 --> 00:26:20.068 align:middle line:90% AUDIENCE: I think I [INAUDIBLE]. 00:26:20.068 --> 00:26:21.110 align:middle line:90% ISAAC SAMUEL KOHANE: 100? 00:26:21.110 --> 00:26:22.235 align:middle line:90% AUDIENCE: No, less than $1. 00:26:22.235 --> 00:26:23.652 align:middle line:84% ISAAC SAMUEL KOHANE: Less than $1. 00:26:23.652 --> 00:26:25.250 align:middle line:90% So let's give you reality. 00:26:25.250 --> 00:26:27.080 align:middle line:90% Today, you'd be charged-- 00:26:27.080 --> 00:26:29.090 align:middle line:84% charge, which is not the question I asked you. 00:26:29.090 --> 00:26:33.050 align:middle line:84% You'd be charged approximately $500 to $1,000. 00:26:33.050 --> 00:26:39.080 align:middle line:84% The actual cost if you're running the test in time 00:26:39.080 --> 00:26:45.770 align:middle line:84% of your technician and materials, so the full cost is, 00:26:45.770 --> 00:26:48.560 align:middle line:84% if you're extremely inefficient, $0.50. 00:26:48.560 --> 00:26:52.520 align:middle line:84% If you're a very efficient, $0.10. 00:26:52.520 --> 00:26:54.740 align:middle line:84% So that's important to realize because we are now 00:26:54.740 --> 00:26:59.360 align:middle line:84% today in an emergency room routinely setting off 00:26:59.360 --> 00:27:03.470 align:middle line:84% the SMA 20, which is 20 different things, like sodium 00:27:03.470 --> 00:27:07.520 align:middle line:84% potassium, chloride, carbon dioxide, BUN, creatinine, 00:27:07.520 --> 00:27:08.720 align:middle line:90% and a few other tests. 00:27:08.720 --> 00:27:12.330 align:middle line:84% And you're charged on the order of-- 00:27:12.330 --> 00:27:14.480 align:middle line:84% or your insurance, hopefully is charged on order 00:27:14.480 --> 00:27:17.840 align:middle line:90% of 100 to 200 bucks to do that. 00:27:17.840 --> 00:27:20.900 align:middle line:84% The marginal cost of this is no higher than 00:27:20.900 --> 00:27:22.160 align:middle line:90% of doing those electrolytes. 00:27:22.160 --> 00:27:25.190 align:middle line:84% So what this tells us is that today, 00:27:25.190 --> 00:27:31.280 align:middle line:84% it's possible to run, for uncommon diseases, 1,000 00:27:31.280 --> 00:27:36.060 align:middle line:84% different snip tests at a cost effective way. 00:27:36.060 --> 00:27:40.022 align:middle line:84% So even though your polymorphism is rare, 00:27:40.022 --> 00:27:41.480 align:middle line:84% if it has clinical significance, we 00:27:41.480 --> 00:27:44.900 align:middle line:84% can find 1,000 different rare polymorphisms 00:27:44.900 --> 00:27:48.860 align:middle line:84% that in aggregate may end up being a lot of people. 00:27:48.860 --> 00:27:51.830 align:middle line:84% And I am not the only one to have recognized this, 00:27:51.830 --> 00:27:54.660 align:middle line:84% and there are now a bunch of companies who understand this. 00:27:54.660 --> 00:27:59.270 align:middle line:84% So the companies who previously were charging $1,000 per tests 00:27:59.270 --> 00:28:02.470 align:middle line:84% are under some cost pressure, and to give you a sort 00:28:02.470 --> 00:28:05.240 align:middle line:84% of reality-based feeling for this, 00:28:05.240 --> 00:28:09.320 align:middle line:84% probably what you'll see over the next 5 to 10 years as this 00:28:09.320 --> 00:28:10.040 align:middle line:90% technology, as-- 00:28:10.040 --> 00:28:12.800 align:middle line:84% understand, this technology is actually a commodity level-- 00:28:12.800 --> 00:28:15.830 align:middle line:84% is that price is going to be crushed down 00:28:15.830 --> 00:28:17.307 align:middle line:90% from the 1,000 level to a 10. 00:28:17.307 --> 00:28:19.640 align:middle line:84% Of course, they're going to resist every step of the way 00:28:19.640 --> 00:28:22.310 align:middle line:84% and try charge every last dollar along the way, 00:28:22.310 --> 00:28:26.030 align:middle line:84% but it's sort of an internet kind of effect 00:28:26.030 --> 00:28:28.070 align:middle line:84% because the true costs are trivial, 00:28:28.070 --> 00:28:31.580 align:middle line:84% and we're charging the government these bulk rate 00:28:31.580 --> 00:28:35.280 align:middle line:84% prices all the time for our human association studies. 00:28:35.280 --> 00:28:37.820 align:middle line:84% So this is a hugely important understanding to have. 00:28:37.820 --> 00:28:39.890 align:middle line:84% And why is this so important to understand? 00:28:39.890 --> 00:28:43.190 align:middle line:84% Because if you look at the number of publications 00:28:43.190 --> 00:28:46.760 align:middle line:84% of different polymorphisms that occurred with time, 00:28:46.760 --> 00:28:47.960 align:middle line:90% this is 1980. 00:28:47.960 --> 00:28:50.210 align:middle line:90% This is 2005. 00:28:50.210 --> 00:28:52.310 align:middle line:84% And what you see here is that it's 00:28:52.310 --> 00:28:57.590 align:middle line:84% been growing steadily and not so slowly every year 00:28:57.590 --> 00:29:01.610 align:middle line:84% so that we're now close to 8,000 publications per year 00:29:01.610 --> 00:29:04.160 align:middle line:84% about reporting on different polymorphisms. 00:29:04.160 --> 00:29:05.810 align:middle line:84% Now as you'll hear on two or three 00:29:05.810 --> 00:29:12.240 align:middle line:84% lectures from now, well over 50% of these are wrong reports, 00:29:12.240 --> 00:29:14.450 align:middle line:84% and one of the things you have to learn in this class 00:29:14.450 --> 00:29:17.750 align:middle line:84% is what constitutes a good versus a bad report, 00:29:17.750 --> 00:29:22.603 align:middle line:84% a protective factor based on a polymorphism. 00:29:22.603 --> 00:29:24.770 align:middle line:84% Nonetheless, it's telling you that there's thousands 00:29:24.770 --> 00:29:27.020 align:middle line:90% of polymorphisms out there. 00:29:27.020 --> 00:29:29.600 align:middle line:84% Even if 50% of the reports are wrong, 00:29:29.600 --> 00:29:31.580 align:middle line:84% thousands of polymorphisms are associated 00:29:31.580 --> 00:29:33.650 align:middle line:84% with clinical disease that are being reported 00:29:33.650 --> 00:29:35.250 align:middle line:90% on literature per year. 00:29:35.250 --> 00:29:36.275 align:middle line:90% So that's a huge number. 00:29:36.275 --> 00:29:38.810 align:middle line:90% 00:29:38.810 --> 00:29:42.410 align:middle line:84% And as we'll get to shortly, genomics 00:29:42.410 --> 00:29:43.850 align:middle line:84% is not about single gene diseases. 00:29:43.850 --> 00:29:46.640 align:middle line:84% But nonetheless, the number of diseases 00:29:46.640 --> 00:29:50.030 align:middle line:84% that are being found to have a monogenic basis 00:29:50.030 --> 00:29:55.430 align:middle line:84% is growing even faster exponentially from 1981 00:29:55.430 --> 00:29:58.010 align:middle line:90% till the Millennium. 00:29:58.010 --> 00:30:02.010 align:middle line:90% 00:30:02.010 --> 00:30:07.770 align:middle line:84% And so what I told you until now the $0.05 figure was, 00:30:07.770 --> 00:30:10.560 align:middle line:84% I wanted to make sure you understood that was true today. 00:30:10.560 --> 00:30:13.050 align:middle line:84% The following is not in the so distant future. 00:30:13.050 --> 00:30:19.590 align:middle line:84% It's actually doable now, but it's at a research basis only. 00:30:19.590 --> 00:30:22.980 align:middle line:90% This is a big chip. 00:30:22.980 --> 00:30:25.585 align:middle line:84% It's a chip made out of many Affymetrix chips sold 00:30:25.585 --> 00:30:27.210 align:middle line:84% by a company called [INAUDIBLE],, which 00:30:27.210 --> 00:30:29.660 align:middle line:84% is a wholly owned subsidiary of Affymetrix. 00:30:29.660 --> 00:30:32.070 align:middle line:84% And I'll tell you more about this chip 00:30:32.070 --> 00:30:33.640 align:middle line:90% two or three lectures from now. 00:30:33.640 --> 00:30:36.810 align:middle line:84% But suffice it to say that instead of interrogating 00:30:36.810 --> 00:30:38.820 align:middle line:84% for a different gene in every spot, 00:30:38.820 --> 00:30:42.300 align:middle line:84% it's interrogating every single possible misspelling 00:30:42.300 --> 00:30:45.590 align:middle line:84% of the entire human genome or-- and I think this is-- yeah, 00:30:45.590 --> 00:30:49.890 align:middle line:84% the entire human genome on this collection of chips so that 00:30:49.890 --> 00:30:52.130 align:middle line:90% in one fell swoop-- 00:30:52.130 --> 00:30:53.970 align:middle line:84% this is actually just for chromosome 21-- 00:30:53.970 --> 00:30:56.970 align:middle line:84% you can actually fully genotype the entire chromosome 00:30:56.970 --> 00:31:01.460 align:middle line:84% of an individual within two hours. 00:31:01.460 --> 00:31:02.930 align:middle line:90% And that's available today. 00:31:02.930 --> 00:31:04.620 align:middle line:90% It's commodity technology. 00:31:04.620 --> 00:31:07.040 align:middle line:84% It's just too expensive to be able to give you 00:31:07.040 --> 00:31:11.480 align:middle line:84% the same sense of reality as I did for single SNP genotyping. 00:31:11.480 --> 00:31:15.940 align:middle line:90% 00:31:15.940 --> 00:31:18.640 align:middle line:84% And I mean, this just shows that they're 00:31:18.640 --> 00:31:21.760 align:middle line:84% able to comprehensively show the frequencies of distribution 00:31:21.760 --> 00:31:28.220 align:middle line:84% of different SNPs across these individuals in one fell swoop. 00:31:28.220 --> 00:31:33.250 align:middle line:84% And that reminds me within the next day, 00:31:33.250 --> 00:31:35.732 align:middle line:84% I'm going to start uploading to the My Courses website-- 00:31:35.732 --> 00:31:36.940 align:middle line:90% I'll get to you in a second-- 00:31:36.940 --> 00:31:41.500 align:middle line:84% to the My Courses website PDFs of all the class reading. 00:31:41.500 --> 00:31:44.110 align:middle line:84% So today, I'm going to backfill it 00:31:44.110 --> 00:31:47.500 align:middle line:84% with things that would be relevant to today's lecture. 00:31:47.500 --> 00:31:51.140 align:middle line:84% Nonetheless, for future lectures, 00:31:51.140 --> 00:31:55.030 align:middle line:84% you will have the PDFs for the lecture of that day 00:31:55.030 --> 00:31:56.890 align:middle line:90% prior to that day. 00:31:56.890 --> 00:31:59.870 align:middle line:90% 00:31:59.870 --> 00:32:03.600 align:middle line:84% So-- you're going to ask a question. 00:32:03.600 --> 00:32:08.060 align:middle line:84% AUDIENCE: Yeah, [INAUDIBLE] do it [INAUDIBLE].. 00:32:08.060 --> 00:32:12.650 align:middle line:84% So these are just identifying a sequence of genes, 00:32:12.650 --> 00:32:17.300 align:middle line:84% or is this coming out of the genome project, 00:32:17.300 --> 00:32:21.300 align:middle line:90% and it's based upon [INAUDIBLE]? 00:32:21.300 --> 00:32:23.000 align:middle line:84% ISAAC SAMUEL KOHANE: So what it's 00:32:23.000 --> 00:32:26.660 align:middle line:84% based is it's a confluence of several industrial-- 00:32:26.660 --> 00:32:28.040 align:middle line:90% well, what is it? 00:32:28.040 --> 00:32:34.312 align:middle line:84% Each little sequence on this microarray corresponds-- 00:32:34.312 --> 00:32:35.270 align:middle line:90% name on the blackboard. 00:32:35.270 --> 00:32:39.470 align:middle line:84% Each sequence corresponds to a stretch of DNA, 00:32:39.470 --> 00:32:41.460 align:middle line:84% and the center of that little stretch of DNA 00:32:41.460 --> 00:32:46.550 align:middle line:84% that stretch of all of the nucleotide, 20 bases. 00:32:46.550 --> 00:32:52.100 align:middle line:84% In the middle of that is the polymorphism. 00:32:52.100 --> 00:32:56.000 align:middle line:84% And so let's say we want to capture all four 00:32:56.000 --> 00:32:57.200 align:middle line:90% possible polymorphisms. 00:32:57.200 --> 00:32:59.780 align:middle line:84% For that stretch of DNA, you'll have 00:32:59.780 --> 00:33:02.690 align:middle line:84% four nucleotides corresponding to the ACE or C 00:33:02.690 --> 00:33:07.020 align:middle line:84% or T or G that's different on that central base. 00:33:07.020 --> 00:33:11.030 align:middle line:84% So on here are all the different polymorphisms 00:33:11.030 --> 00:33:14.340 align:middle line:84% that are known for chromosome 21. 00:33:14.340 --> 00:33:15.890 align:middle line:90% So how do they get that? 00:33:15.890 --> 00:33:18.290 align:middle line:84% Is a combination of the verified SNPs 00:33:18.290 --> 00:33:20.960 align:middle line:84% from the Human Genome Project, the verified 00:33:20.960 --> 00:33:23.907 align:middle line:84% SNPs from Celera which they had to pay 00:33:23.907 --> 00:33:25.490 align:middle line:84% cold cash for-- but they're a company, 00:33:25.490 --> 00:33:26.782 align:middle line:90% they have no problem with that. 00:33:26.782 --> 00:33:30.470 align:middle line:84% And they also licensed SNPs from ABI, 00:33:30.470 --> 00:33:33.770 align:middle line:90% from [INAUDIBLE] Biosystems. 00:33:33.770 --> 00:33:39.060 align:middle line:84% And finally, they did their own resequencing 00:33:39.060 --> 00:33:45.110 align:middle line:84% so they would know what are all the SNPs, the common SNPs, 00:33:45.110 --> 00:33:46.770 align:middle line:90% for this region. 00:33:46.770 --> 00:33:49.070 align:middle line:84% So you're able to, in one fell swoop, 00:33:49.070 --> 00:33:52.520 align:middle line:84% directly assess through hybridization of DNA now, 00:33:52.520 --> 00:33:57.590 align:middle line:84% not RNA, which are the genotypes of the ones that are known. 00:33:57.590 --> 00:34:04.950 align:middle line:84% Now if it was not known, then it wouldn't show up here. 00:34:04.950 --> 00:34:07.590 align:middle line:84% AUDIENCE: So just as a follow up question, 00:34:07.590 --> 00:34:12.469 align:middle line:84% seeing as how these are SNPs that are known and taken 00:34:12.469 --> 00:34:16.429 align:middle line:84% from the sources and we know that those sources are 00:34:16.429 --> 00:34:18.278 align:middle line:90% basically a few sample subjects. 00:34:18.278 --> 00:34:19.320 align:middle line:90% ISAAC SAMUEL KOHANE: Yes. 00:34:19.320 --> 00:34:21.502 align:middle line:84% AUDIENCE: --are we going to talk at all 00:34:21.502 --> 00:34:22.989 align:middle line:90% in the course [INAUDIBLE]. 00:34:22.989 --> 00:34:24.114 align:middle line:90% ISAAC SAMUEL KOHANE: Class. 00:34:24.114 --> 00:34:27.610 align:middle line:84% AUDIENCE: --the class about the implications 00:34:27.610 --> 00:34:31.670 align:middle line:84% of a very small pool of individuals 00:34:31.670 --> 00:34:34.110 align:middle line:84% an compare with other people's [INAUDIBLE]?? 00:34:34.110 --> 00:34:36.110 align:middle line:84% ISAAC SAMUEL KOHANE: We are definitely-- so what 00:34:36.110 --> 00:34:39.440 align:middle line:84% you're referring to is, for instance, the Human Genome 00:34:39.440 --> 00:34:42.120 align:middle line:84% Project and the Celera project were based on 00:34:42.120 --> 00:34:44.060 align:middle line:90% well under 20 individuals each. 00:34:44.060 --> 00:34:47.719 align:middle line:90% And so I have two columns. 00:34:47.719 --> 00:34:49.969 align:middle line:84% One, a column which I'm not too comfortable with 00:34:49.969 --> 00:34:54.080 align:middle line:84% but nonetheless true is that if it's a common SNP, 00:34:54.080 --> 00:35:00.890 align:middle line:84% it'll show up typically in one of those 20 more or less. 00:35:00.890 --> 00:35:02.960 align:middle line:84% The danger is the opposite, is that a lot 00:35:02.960 --> 00:35:06.830 align:middle line:84% of private mutations-- in other words, you may share a SNP with 00:35:06.830 --> 00:35:09.890 align:middle line:84% and your immediate family, wherever you came from, 00:35:09.890 --> 00:35:11.630 align:middle line:90% that's not present in anybody. 00:35:11.630 --> 00:35:14.030 align:middle line:84% But if you're one of the 20, that now 00:35:14.030 --> 00:35:16.250 align:middle line:84% has this major representation in this database, 00:35:16.250 --> 00:35:18.953 align:middle line:84% and it's probably the other direction 00:35:18.953 --> 00:35:20.120 align:middle line:90% that we have to worry about. 00:35:20.120 --> 00:35:22.670 align:middle line:84% Nonetheless the answer to it is very simply this-- 00:35:22.670 --> 00:35:26.000 align:middle line:84% is massive resequencing of a lot more individuals. 00:35:26.000 --> 00:35:28.070 align:middle line:84% And as this technology becomes commoditized, 00:35:28.070 --> 00:35:30.960 align:middle line:84% that problem is going to go away. 00:35:30.960 --> 00:35:34.130 align:middle line:84% So the HapMap map project, for instance, 00:35:34.130 --> 00:35:36.680 align:middle line:84% is a project, which you'll hear about during this class. 00:35:36.680 --> 00:35:40.430 align:middle line:84% And it's a project to actually define the structure of SNPs 00:35:40.430 --> 00:35:45.110 align:middle line:84% across larger populations and across ethnically diverse 00:35:45.110 --> 00:35:47.840 align:middle line:84% populations because as I hinted to you in that term 00:35:47.840 --> 00:35:49.790 align:middle line:84% private mutations, you can imagine 00:35:49.790 --> 00:35:52.820 align:middle line:84% or you will be taught that there are different subpopulations 00:35:52.820 --> 00:35:54.710 align:middle line:84% that have these SNPs that are only 00:35:54.710 --> 00:35:57.860 align:middle line:84% present in a subpopulation either within a subpopulation 00:35:57.860 --> 00:35:59.548 align:middle line:84% called your family or a subpopulation 00:35:59.548 --> 00:36:01.340 align:middle line:84% called people that came from that continent 00:36:01.340 --> 00:36:03.330 align:middle line:90% that you originally came from. 00:36:03.330 --> 00:36:06.487 align:middle line:84% So the short answer is it's a problem, 00:36:06.487 --> 00:36:08.570 align:middle line:84% and it's a problem more on one side and the other. 00:36:08.570 --> 00:36:11.595 align:middle line:90% In other words, since-- 00:36:11.595 --> 00:36:12.470 align:middle line:90% let me rephrase that. 00:36:12.470 --> 00:36:16.310 align:middle line:84% There's a certain frequency of common SNPs, about one 00:36:16.310 --> 00:36:18.600 align:middle line:90% per kilo base. 00:36:18.600 --> 00:36:20.720 align:middle line:84% But if you look at rare SNPs, they 00:36:20.720 --> 00:36:22.680 align:middle line:84% have a much, much higher frequency. 00:36:22.680 --> 00:36:24.740 align:middle line:84% In other words, I have a bunch of SNPs 00:36:24.740 --> 00:36:30.410 align:middle line:84% that are just unique to me and a few thousand people that 00:36:30.410 --> 00:36:32.450 align:middle line:90% have a shared heritage with me. 00:36:32.450 --> 00:36:34.350 align:middle line:84% And those probably don't have, by the way, 00:36:34.350 --> 00:36:36.500 align:middle line:84% for the most part, any clinical significance. 00:36:36.500 --> 00:36:38.600 align:middle line:84% But if I were one of the 20, it'd 00:36:38.600 --> 00:36:40.652 align:middle line:84% be over-represented in that sample 00:36:40.652 --> 00:36:41.735 align:middle line:90% if it's not a common snip. 00:36:41.735 --> 00:36:44.876 align:middle line:90% 00:36:44.876 --> 00:36:45.840 align:middle line:90% Ah. 00:36:45.840 --> 00:36:50.400 align:middle line:84% So this makes the following simple point. 00:36:50.400 --> 00:36:52.960 align:middle line:84% Why-- it's answering the following question. 00:36:52.960 --> 00:36:55.800 align:middle line:84% Why is computation such a central part 00:36:55.800 --> 00:36:57.720 align:middle line:90% to genomic medicine? 00:36:57.720 --> 00:37:04.440 align:middle line:84% And simply this-- these two 1999 style stock 00:37:04.440 --> 00:37:08.280 align:middle line:84% curves, stock market curves, correspond to the following two 00:37:08.280 --> 00:37:08.970 align:middle line:90% processes. 00:37:08.970 --> 00:37:11.580 align:middle line:84% This shallower curve is the number 00:37:11.580 --> 00:37:14.940 align:middle line:84% of publications in "Medline," the online repository 00:37:14.940 --> 00:37:16.980 align:middle line:84% of articles, and goes all the way back 00:37:16.980 --> 00:37:19.320 align:middle line:90% to a century, Index Medicus. 00:37:19.320 --> 00:37:21.820 align:middle line:84% And it's [INAUDIBLE],, say, exponential curve. 00:37:21.820 --> 00:37:23.670 align:middle line:84% But because of the industrialization 00:37:23.670 --> 00:37:26.100 align:middle line:84% of gene sequencing, the number of sequences 00:37:26.100 --> 00:37:30.250 align:middle line:84% have risen at a far higher exponential curve. 00:37:30.250 --> 00:37:35.950 align:middle line:84% And what you can think of this, as a proxy for the knowledge 00:37:35.950 --> 00:37:36.450 align:middle line:90% gap. 00:37:36.450 --> 00:37:38.370 align:middle line:84% If you take this to be genomic data, 00:37:38.370 --> 00:37:41.520 align:middle line:84% the genome sequences being genomic data and this being 00:37:41.520 --> 00:37:44.070 align:middle line:84% as a proxy for our knowledge, this gap is. 00:37:44.070 --> 00:37:46.430 align:middle line:84% Growing and it's growing even faster 00:37:46.430 --> 00:37:48.930 align:middle line:84% than it's portrayed here because it's not just DNA sequence. 00:37:48.930 --> 00:37:51.540 align:middle line:84% It's genetic maps physical, maps, polymorphism, 00:37:51.540 --> 00:37:54.270 align:middle line:84% structure information, gene transcription patterns, 00:37:54.270 --> 00:37:55.710 align:middle line:90% protein translation activity. 00:37:55.710 --> 00:37:58.390 align:middle line:84% And all of these end up in databases. 00:37:58.390 --> 00:38:02.190 align:middle line:84% And if you are like me in the middle of a Medical Center, 00:38:02.190 --> 00:38:04.620 align:middle line:84% You'll find that there's very, very few people who 00:38:04.620 --> 00:38:07.260 align:middle line:84% are trained as clinicians and who 00:38:07.260 --> 00:38:10.050 align:middle line:84% have the requisite computational skills. 00:38:10.050 --> 00:38:11.880 align:middle line:84% So I have a PhD in computer science, 00:38:11.880 --> 00:38:14.183 align:middle line:84% but that's not the typical medical path 00:38:14.183 --> 00:38:15.600 align:middle line:84% a lot of you in this room actually 00:38:15.600 --> 00:38:20.070 align:middle line:84% have the nice joint skill set, but you're in the minority. 00:38:20.070 --> 00:38:23.280 align:middle line:84% And consequently, because all these data types 00:38:23.280 --> 00:38:27.450 align:middle line:84% end up in a database, trying to translate this into medicine 00:38:27.450 --> 00:38:32.040 align:middle line:84% is fundamentally devolved to the responsibility of individuals 00:38:32.040 --> 00:38:35.010 align:middle line:84% who can both understand the biological problems 00:38:35.010 --> 00:38:41.010 align:middle line:84% and translate through a variety of computational techniques 00:38:41.010 --> 00:38:42.850 align:middle line:84% all this data into more knowledge. 00:38:42.850 --> 00:38:46.470 align:middle line:84% So the hubris, the conceit of [? bioinformaticians ?] 00:38:46.470 --> 00:38:48.690 align:middle line:84% is that somehow we can use computational techniques 00:38:48.690 --> 00:38:51.030 align:middle line:90% to raise this to that level. 00:38:51.030 --> 00:38:53.820 align:middle line:90% 00:38:53.820 --> 00:38:55.560 align:middle line:90% We did the [INAUDIBLE] already. 00:38:55.560 --> 00:38:57.238 align:middle line:84% So now let's talk about what is genomic 00:38:57.238 --> 00:38:59.280 align:middle line:84% versus genetic because you hear these terms a lot 00:38:59.280 --> 00:39:01.080 align:middle line:90% as if they're interchangeable. 00:39:01.080 --> 00:39:04.080 align:middle line:84% And they kind of will be, but they're not quite yet. 00:39:04.080 --> 00:39:05.680 align:middle line:84% And it depends on your perspective. 00:39:05.680 --> 00:39:08.512 align:middle line:90% So let me just tell this joke. 00:39:08.512 --> 00:39:09.720 align:middle line:90% I hope you think it's a joke. 00:39:09.720 --> 00:39:12.790 align:middle line:84% An engineer, a physicist, a mathematician, 00:39:12.790 --> 00:39:14.850 align:middle line:84% a computer scientist, and a statistician 00:39:14.850 --> 00:39:17.970 align:middle line:84% are on a train heading north and had just crossed the border 00:39:17.970 --> 00:39:18.815 align:middle line:90% into Scotland. 00:39:18.815 --> 00:39:20.940 align:middle line:84% They look out the window and they see a black sheep 00:39:20.940 --> 00:39:23.940 align:middle line:90% for the first time. 00:39:23.940 --> 00:39:26.400 align:middle line:90% The engineer exclaims, look. 00:39:26.400 --> 00:39:27.480 align:middle line:90% Scottish sheep are black. 00:39:27.480 --> 00:39:30.230 align:middle line:90% 00:39:30.230 --> 00:39:34.620 align:middle line:84% The physicist yells, no, no, some Scottish sheep are black. 00:39:34.620 --> 00:39:36.420 align:middle line:84% The mathematician looks irritated and says, 00:39:36.420 --> 00:39:38.280 align:middle line:84% there is at least one field containing 00:39:38.280 --> 00:39:43.170 align:middle line:84% at least one sheep of which at least one side is black. 00:39:43.170 --> 00:39:47.940 align:middle line:84% The computer scientist says, oh no, a special case. 00:39:47.940 --> 00:39:53.760 align:middle line:84% And in fact, it's how you view genomics and genetics actually 00:39:53.760 --> 00:39:56.610 align:middle line:84% corresponds to where you find yourself on this spectrum. 00:39:56.610 --> 00:39:58.157 align:middle line:84% So that will probably inform you why 00:39:58.157 --> 00:40:00.613 align:middle line:90% there is some global confusion. 00:40:00.613 --> 00:40:02.280 align:middle line:84% And finally, the statistician, I forgot, 00:40:02.280 --> 00:40:05.590 align:middle line:84% says it is not statistically significant. 00:40:05.590 --> 00:40:08.540 align:middle line:84% So here's one view of genomic versus genetic. 00:40:08.540 --> 00:40:10.620 align:middle line:84% So genetic medicine, the kind of medicine 00:40:10.620 --> 00:40:12.720 align:middle line:84% that's been around for us for a long time, 00:40:12.720 --> 00:40:16.920 align:middle line:84% has to do with low frequencies of high penetrance genes 00:40:16.920 --> 00:40:18.570 align:middle line:84% by which I mean genes that if you 00:40:18.570 --> 00:40:22.140 align:middle line:84% have that particular mutation in that gene 00:40:22.140 --> 00:40:24.420 align:middle line:84% is going to give you a high likelihood of having 00:40:24.420 --> 00:40:25.650 align:middle line:90% that disease. 00:40:25.650 --> 00:40:28.690 align:middle line:84% And there's thousands of these relatively uncommon diseases. 00:40:28.690 --> 00:40:31.380 align:middle line:84% The most common of these, hemochromatosis, 00:40:31.380 --> 00:40:33.570 align:middle line:84% which is iron deposition in the liver, 00:40:33.570 --> 00:40:36.960 align:middle line:84% has a frequency in the population about 1 to 300. 00:40:36.960 --> 00:40:39.810 align:middle line:84% And these diseases previously had been mostly 00:40:39.810 --> 00:40:42.540 align:middle line:84% assessed indirectly or focused on single genes 00:40:42.540 --> 00:40:44.790 align:middle line:90% by which indirectly linkage-- 00:40:44.790 --> 00:40:46.740 align:middle line:84% for instance, linkage within a family. 00:40:46.740 --> 00:40:51.420 align:middle line:84% In contrast, genomic medicine is about not 00:40:51.420 --> 00:40:53.640 align:middle line:84% in a rare disease but just common diseases. 00:40:53.640 --> 00:41:00.480 align:middle line:84% It's diabetes, cancer, heart disease, inflammatory diseases. 00:41:00.480 --> 00:41:03.235 align:middle line:84% And therefore, almost by definition, 00:41:03.235 --> 00:41:04.860 align:middle line:84% the geniculate risk for common diseases 00:41:04.860 --> 00:41:07.740 align:middle line:84% will often due the disease producing areas 00:41:07.740 --> 00:41:09.420 align:middle line:84% with relatively high frequencies-- that 00:41:09.420 --> 00:41:11.010 align:middle line:90% is, greater than 1%. 00:41:11.010 --> 00:41:13.620 align:middle line:84% And all these genes in this perspective 00:41:13.620 --> 00:41:15.300 align:middle line:90% may be disease-causing. 00:41:15.300 --> 00:41:17.702 align:middle line:84% And they're not for uncommon diseases. 00:41:17.702 --> 00:41:19.410 align:middle line:84% They're for common disorders, and they're 00:41:19.410 --> 00:41:21.180 align:middle line:84% due to the interactions of multiple genes 00:41:21.180 --> 00:41:22.870 align:middle line:90% and environmental factors. 00:41:22.870 --> 00:41:25.212 align:middle line:84% And because of the industrialization 00:41:25.212 --> 00:41:27.420 align:middle line:84% in genomic medicine, we do direct experimental access 00:41:27.420 --> 00:41:29.370 align:middle line:84% to the entire genome in the fashion 00:41:29.370 --> 00:41:30.990 align:middle line:90% that I just illustrated. 00:41:30.990 --> 00:41:33.780 align:middle line:84% Someone else, [INAUDIBLE] and [INAUDIBLE],, 00:41:33.780 --> 00:41:36.690 align:middle line:84% have given a related similar distinction 00:41:36.690 --> 00:41:38.490 align:middle line:90% between genetic and genomic. 00:41:38.490 --> 00:41:40.920 align:middle line:84% In genetics, you're looking at basically a structure 00:41:40.920 --> 00:41:43.302 align:middle line:90% of genes, the sequence. 00:41:43.302 --> 00:41:44.760 align:middle line:84% In genomic medicine, you're looking 00:41:44.760 --> 00:41:46.650 align:middle line:90% at the function of these genes. 00:41:46.650 --> 00:41:51.060 align:middle line:84% In genetic disease, you're looking at genomics. 00:41:51.060 --> 00:41:52.110 align:middle line:90% You're looking at DNA. 00:41:52.110 --> 00:41:54.090 align:middle line:84% In genomic medicine, we're looking 00:41:54.090 --> 00:41:55.680 align:middle line:90% at more than just the gene. 00:41:55.680 --> 00:41:59.290 align:middle line:84% We're looking at the RNA and the proteins. 00:41:59.290 --> 00:42:00.870 align:middle line:84% This is what I referred to already. 00:42:00.870 --> 00:42:02.880 align:middle line:84% Map-based gene discovery-- in other words, 00:42:02.880 --> 00:42:07.015 align:middle line:84% probabilistic linkage between loci versus direct sequence 00:42:07.015 --> 00:42:07.890 align:middle line:90% based gene discovery. 00:42:07.890 --> 00:42:12.457 align:middle line:84% I mean, now, diseases are being discovered computationally. 00:42:12.457 --> 00:42:14.040 align:middle line:84% I don't know if I'll get around to it, 00:42:14.040 --> 00:42:16.890 align:middle line:84% but there's a great example now of how people are discovering 00:42:16.890 --> 00:42:20.970 align:middle line:84% cholesterol risk factor genes purely based 00:42:20.970 --> 00:42:26.340 align:middle line:84% on our shared attributes in our genome between man and mouse. 00:42:26.340 --> 00:42:29.280 align:middle line:84% Without having ever poured a beaker, 00:42:29.280 --> 00:42:31.710 align:middle line:84% people are actually being able to computationally 00:42:31.710 --> 00:42:34.210 align:middle line:84% just identify new risk factors in that fashion. 00:42:34.210 --> 00:42:35.880 align:middle line:90% But perhaps I will get to that. 00:42:35.880 --> 00:42:39.270 align:middle line:84% It just occurred to me that I don't have lecture anymore 00:42:39.270 --> 00:42:43.410 align:middle line:84% on comparative genomics, which is how do we actually 00:42:43.410 --> 00:42:46.170 align:middle line:84% take advantage of the genomes of other species 00:42:46.170 --> 00:42:47.730 align:middle line:84% to inform us about human medicine. 00:42:47.730 --> 00:42:51.510 align:middle line:84% Maybe I should retroactively fit that in. 00:42:51.510 --> 00:42:54.420 align:middle line:84% And again, in genetics, it's monogenic disorders. 00:42:54.420 --> 00:42:57.660 align:middle line:84% In genomics, it's multifactorial disorders. 00:42:57.660 --> 00:43:01.230 align:middle line:84% And there's a specific DNA diagnosis in genetic disease. 00:43:01.230 --> 00:43:02.280 align:middle line:90% You have this disease. 00:43:02.280 --> 00:43:04.500 align:middle line:84% As opposed to genomics, it's more 00:43:04.500 --> 00:43:08.100 align:middle line:84% you're at risk for the following factors because it's not, 00:43:08.100 --> 00:43:11.460 align:middle line:84% in fact, a high penetrance disease. 00:43:11.460 --> 00:43:14.430 align:middle line:84% And looking at gene action and genetic disease, 00:43:14.430 --> 00:43:17.910 align:middle line:84% and you're looking more at genetic regulation 00:43:17.910 --> 00:43:19.410 align:middle line:90% in genetic medicine. 00:43:19.410 --> 00:43:22.950 align:middle line:84% And etiology, specific mutation versus-- 00:43:22.950 --> 00:43:24.840 align:middle line:84% because a mutation just tells you there's 00:43:24.840 --> 00:43:28.088 align:middle line:84% an error which is associated with a disease 00:43:28.088 --> 00:43:29.880 align:middle line:84% whereas in genomic medicine, you're saying, 00:43:29.880 --> 00:43:33.780 align:middle line:84% how did this error come to give rise to this disease? 00:43:33.780 --> 00:43:36.330 align:middle line:84% And in genetics, it's one species. 00:43:36.330 --> 00:43:37.890 align:middle line:84% In genomics, it's several species. 00:43:37.890 --> 00:43:40.680 align:middle line:84% Now I think you could in good faith 00:43:40.680 --> 00:43:43.110 align:middle line:84% actually argue that there's a lot of overlap in these two, 00:43:43.110 --> 00:43:47.340 align:middle line:84% but that's broadly how the world seems to be divided. 00:43:47.340 --> 00:43:50.710 align:middle line:84% And I believe that will be blurred soon. 00:43:50.710 --> 00:43:52.530 align:middle line:84% But I think the important take-home message 00:43:52.530 --> 00:43:57.330 align:middle line:84% is that in genomic medicine, we really look comprehensively 00:43:57.330 --> 00:43:59.050 align:middle line:90% at all data types. 00:43:59.050 --> 00:44:00.930 align:middle line:84% And basically, you view the patient, 00:44:00.930 --> 00:44:03.570 align:middle line:84% if you want to look at them in a reductionist fashion, 00:44:03.570 --> 00:44:05.280 align:middle line:90% as the following kind of matrix. 00:44:05.280 --> 00:44:08.940 align:middle line:84% We're looking at environmental data, single nucleotide 00:44:08.940 --> 00:44:11.280 align:middle line:84% polymorphisms, proteomic measurements, gene expression 00:44:11.280 --> 00:44:13.260 align:middle line:84% measurements, and clinical measurements 00:44:13.260 --> 00:44:16.410 align:middle line:84% such as your history, your physical exam, laboratory 00:44:16.410 --> 00:44:19.740 align:middle line:84% studies, and imaging studies all across time. 00:44:19.740 --> 00:44:21.720 align:middle line:84% But although there is missing data 00:44:21.720 --> 00:44:23.160 align:middle line:84% and although these are incomplete, 00:44:23.160 --> 00:44:26.700 align:middle line:84% we can now view the patient as this big time series 00:44:26.700 --> 00:44:29.490 align:middle line:84% that you can apply a lot of machine learning techniques 00:44:29.490 --> 00:44:33.150 align:middle line:84% to cluster patients together, create new diagnosis 00:44:33.150 --> 00:44:36.750 align:middle line:84% and predictive modeling techniques 00:44:36.750 --> 00:44:40.170 align:middle line:84% to figure out where in this space of possibilities 00:44:40.170 --> 00:44:42.150 align:middle line:84% of these attributes is the patient 00:44:42.150 --> 00:44:46.120 align:middle line:90% likely to be in the future. 00:44:46.120 --> 00:44:48.720 align:middle line:84% And it's this comprehensiveness that I think 00:44:48.720 --> 00:44:50.100 align:middle line:90% characterizes genomic medicine. 00:44:50.100 --> 00:44:52.870 align:middle line:84% As we will get to later in the course, 00:44:52.870 --> 00:44:56.040 align:middle line:84% I will identify how unfortunately, this 00:44:56.040 --> 00:44:57.930 align:middle line:90% is becoming a commodity. 00:44:57.930 --> 00:45:00.150 align:middle line:84% Only genomic measurements are cheap, 00:45:00.150 --> 00:45:01.500 align:middle line:90% as I just illustrated you. 00:45:01.500 --> 00:45:03.270 align:middle line:84% But this part, because we have to deal 00:45:03.270 --> 00:45:06.510 align:middle line:84% with these dumb doctors, it's very expensive 00:45:06.510 --> 00:45:07.800 align:middle line:90% because it's labor intensive. 00:45:07.800 --> 00:45:10.290 align:middle line:84% Trying to assess the phenotype of the patient 00:45:10.290 --> 00:45:13.680 align:middle line:84% is a labor intensive process just as medical care is, just 00:45:13.680 --> 00:45:14.790 align:middle line:90% as education is. 00:45:14.790 --> 00:45:17.710 align:middle line:90% 00:45:17.710 --> 00:45:22.000 align:middle line:84% And so why do we have to be comprehensive? 00:45:22.000 --> 00:45:26.200 align:middle line:84% Other than just liking to be comprehensive, 00:45:26.200 --> 00:45:27.460 align:middle line:90% what is our motivation? 00:45:27.460 --> 00:45:29.510 align:middle line:90% And the answer is very simple. 00:45:29.510 --> 00:45:33.610 align:middle line:84% It comes from an understanding that the phenotype is not 00:45:33.610 --> 00:45:37.390 align:middle line:84% just directly just probabilistically linked 00:45:37.390 --> 00:45:39.040 align:middle line:90% to some gene. 00:45:39.040 --> 00:45:44.890 align:middle line:84% So in the old genetic medicine, you 00:45:44.890 --> 00:45:49.160 align:middle line:84% would be looking at some gene, some marker, some polymorphism 00:45:49.160 --> 00:45:51.640 align:middle line:84% on some gene, and say, ah, this gene 00:45:51.640 --> 00:45:55.300 align:middle line:84% has a G in position 291, whenever this person has 00:45:55.300 --> 00:45:56.770 align:middle line:90% a risk for heart disease. 00:45:56.770 --> 00:46:00.790 align:middle line:84% And what you'd really be saying is when this gene has 00:46:00.790 --> 00:46:03.352 align:middle line:84% this spelling, it's associated probabilistically 00:46:03.352 --> 00:46:04.810 align:middle line:84% with some other gene that you don't 00:46:04.810 --> 00:46:10.760 align:middle line:84% know that is directly causing a change in the phenotype. 00:46:10.760 --> 00:46:13.840 align:middle line:84% So in genomic medicine, we want to measure this gene but also 00:46:13.840 --> 00:46:19.690 align:middle line:84% every other gene that might be contributing to this phenotype 00:46:19.690 --> 00:46:24.430 align:middle line:84% because as we well know, even if you stuck in the same room 00:46:24.430 --> 00:46:29.950 align:middle line:84% the three individuals with the same polymorphism of that gene, 00:46:29.950 --> 00:46:31.930 align:middle line:84% they would actually have different diseases. 00:46:31.930 --> 00:46:33.670 align:middle line:84% If you look at Huntington's disease, 00:46:33.670 --> 00:46:35.860 align:middle line:84% Huntington's disease is this awful disease 00:46:35.860 --> 00:46:38.860 align:middle line:84% where you lose control of your limbs at age 40 00:46:38.860 --> 00:46:44.290 align:middle line:84% and become demented because of an expanding number of [? CG ?] 00:46:44.290 --> 00:46:47.980 align:middle line:84% repeats in the Huntington's gene. 00:46:47.980 --> 00:46:50.730 align:middle line:84% The longer that number repeats, the larger 00:46:50.730 --> 00:46:53.410 align:middle line:84% the number of repeats, the more acute is the onset, 00:46:53.410 --> 00:46:55.190 align:middle line:90% the earlier is the onset. 00:46:55.190 --> 00:46:58.390 align:middle line:84% But if you take on average, if you take a set of individuals 00:46:58.390 --> 00:47:02.200 align:middle line:84% with the same length of repeats and study them, 00:47:02.200 --> 00:47:05.470 align:middle line:84% they have a wide distribution for that given length of repeat 00:47:05.470 --> 00:47:06.970 align:middle line:90% and the onset of disease. 00:47:06.970 --> 00:47:08.180 align:middle line:90% And why is that? 00:47:08.180 --> 00:47:10.957 align:middle line:84% Well, it's because they have a different genetic background, 00:47:10.957 --> 00:47:13.040 align:middle line:84% and these other genes, these other genes involving 00:47:13.040 --> 00:47:15.250 align:middle line:84% the neurotransmitters, structural genes having 00:47:15.250 --> 00:47:20.110 align:middle line:84% to do with the structure of synapses, and so on, 00:47:20.110 --> 00:47:21.530 align:middle line:90% are also influencing it. 00:47:21.530 --> 00:47:25.180 align:middle line:84% And therefore, this particular gene's action on the phenotype 00:47:25.180 --> 00:47:26.980 align:middle line:84% is conditional on all these other genes. 00:47:26.980 --> 00:47:29.230 align:middle line:84% And furthermore, let's not forget 00:47:29.230 --> 00:47:33.340 align:middle line:84% about the common environment and the individual environment 00:47:33.340 --> 00:47:37.640 align:middle line:84% and cultural factors which can further influence this. 00:47:37.640 --> 00:47:39.790 align:middle line:84% And I'm going to shortly point out to you 00:47:39.790 --> 00:47:42.370 align:middle line:90% that this is all too real. 00:47:42.370 --> 00:47:44.080 align:middle line:84% So in genetics, just to remind you, 00:47:44.080 --> 00:47:46.090 align:middle line:84% genetic medicine, basically the study of what 00:47:46.090 --> 00:47:48.730 align:middle line:84% was the probability of a particular phenotype happening 00:47:48.730 --> 00:47:52.900 align:middle line:84% given a particular genotype, and genomic medicine 00:47:52.900 --> 00:47:55.840 align:middle line:84% would try to assess all the genes directly and capture 00:47:55.840 --> 00:47:59.600 align:middle line:84% these other factors and see how they interact. 00:47:59.600 --> 00:48:01.510 align:middle line:84% And this creates a fundamental problem, 00:48:01.510 --> 00:48:03.820 align:middle line:84% and the question you might ask yourself 00:48:03.820 --> 00:48:07.540 align:middle line:84% is, why in genomic medicine are biostatisticians 00:48:07.540 --> 00:48:10.990 align:middle line:84% not front and center as part of the process? 00:48:10.990 --> 00:48:14.050 align:middle line:84% After all, in terms of the quantitative analysis 00:48:14.050 --> 00:48:15.730 align:middle line:84% of clinical outcomes, for instance, 00:48:15.730 --> 00:48:18.610 align:middle line:84% they have been quite omnipresent. 00:48:18.610 --> 00:48:20.308 align:middle line:90% And the answer is simply this. 00:48:20.308 --> 00:48:22.600 align:middle line:84% In a traditional clinical study like the Nurses' Health 00:48:22.600 --> 00:48:24.820 align:middle line:84% Study or the Framingham Heart Study, 00:48:24.820 --> 00:48:28.730 align:middle line:84% you would have on the order of tens of thousands of patients 00:48:28.730 --> 00:48:30.490 align:middle line:84% and at most hundreds of variables. 00:48:30.490 --> 00:48:34.510 align:middle line:84% And there's a lot of tried and true autoregressive models-- 00:48:34.510 --> 00:48:38.590 align:middle line:84% for instance, T-tests, parametric and non-parametric 00:48:38.590 --> 00:48:39.130 align:middle line:90% tests-- 00:48:39.130 --> 00:48:42.800 align:middle line:84% which work extremely well for this kind of data set. 00:48:42.800 --> 00:48:45.260 align:middle line:84% Unfortunately, in genomics, we have quite the opposite. 00:48:45.260 --> 00:48:48.680 align:middle line:84% We have 10,000 or hundreds of thousands 00:48:48.680 --> 00:48:50.980 align:middle line:84% if you proteomics variables, and we only 00:48:50.980 --> 00:48:53.518 align:middle line:84% have, on a good day, hundreds of patients, 00:48:53.518 --> 00:48:55.810 align:middle line:84% especially if you're looking at something in the brain. 00:48:55.810 --> 00:48:57.700 align:middle line:84% As I explained to you at the beginning, 00:48:57.700 --> 00:49:00.075 align:middle line:84% you're only going to get a few hundred brains, let's say, 00:49:00.075 --> 00:49:01.340 align:middle line:90% with brain cancer. 00:49:01.340 --> 00:49:04.360 align:middle line:84% And so let me put this in sort of grade school terms. 00:49:04.360 --> 00:49:06.890 align:middle line:84% If you have one variable, how many questions 00:49:06.890 --> 00:49:09.310 align:middle line:90% do you need to solve it? 00:49:09.310 --> 00:49:09.925 align:middle line:90% one. 00:49:09.925 --> 00:49:11.800 align:middle line:84% If you have two variables, how many equations 00:49:11.800 --> 00:49:12.640 align:middle line:90% do you need to solve it? 00:49:12.640 --> 00:49:13.270 align:middle line:90% Two. 00:49:13.270 --> 00:49:15.280 align:middle line:84% And if you have 10,000, how many do you need to solve it? 00:49:15.280 --> 00:49:15.970 align:middle line:90% A lot. 00:49:15.970 --> 00:49:19.550 align:middle line:84% If you have fewer equations than that, then it's undetermined. 00:49:19.550 --> 00:49:21.700 align:middle line:84% In other words, there's many possible solutions 00:49:21.700 --> 00:49:24.090 align:middle line:84% to the relationships of the variables 00:49:24.090 --> 00:49:25.090 align:middle line:90% than you have available. 00:49:25.090 --> 00:49:27.610 align:middle line:84% And this is so darned undetermined 00:49:27.610 --> 00:49:30.970 align:middle line:84% that all these traditional statistical techniques really 00:49:30.970 --> 00:49:32.380 align:middle line:90% did not work well. 00:49:32.380 --> 00:49:34.870 align:middle line:84% And therefore, frankly, the statistical community 00:49:34.870 --> 00:49:37.180 align:middle line:84% until recently were just not interested in this area. 00:49:37.180 --> 00:49:39.172 align:middle line:84% Whereas the computer science community, 00:49:39.172 --> 00:49:41.380 align:middle line:84% probably because they're just not really smart enough 00:49:41.380 --> 00:49:45.340 align:middle line:84% to understand how difficult a problem this was, just started 00:49:45.340 --> 00:49:47.530 align:middle line:84% applying a bunch of techniques that they had applied 00:49:47.530 --> 00:49:53.200 align:middle line:84% for other high dimensionality data sets such as vision, face 00:49:53.200 --> 00:49:56.560 align:middle line:84% recognition, for instance, and hearing recognition, which also 00:49:56.560 --> 00:49:59.890 align:middle line:84% have a high number of features, applied these very same 00:49:59.890 --> 00:50:01.990 align:middle line:90% techniques to these data sets. 00:50:01.990 --> 00:50:04.630 align:middle line:84% And that just gives you a cultural reason 00:50:04.630 --> 00:50:08.170 align:middle line:84% why computer scientists are the bioinformaticians right now 00:50:08.170 --> 00:50:10.600 align:middle line:84% the center of genetic medicine rather than statisticians, 00:50:10.600 --> 00:50:14.260 align:middle line:84% although they are now coming into play. 00:50:14.260 --> 00:50:17.140 align:middle line:84% So let me give you a feeling for the new pharmacology. 00:50:17.140 --> 00:50:20.260 align:middle line:84% This was a study that I did with a Atul Butte 00:50:20.260 --> 00:50:21.850 align:middle line:90% now four years ago. 00:50:21.850 --> 00:50:25.330 align:middle line:90% 00:50:25.330 --> 00:50:29.590 align:middle line:84% The National Cancer Institute has 60 cancer cell lines 00:50:29.590 --> 00:50:33.530 align:middle line:84% called the NCI 60, which they test for efficacy 00:50:33.530 --> 00:50:35.410 align:middle line:90% a variety of different drugs. 00:50:35.410 --> 00:50:40.630 align:middle line:84% And they have, in their banks, thousands of pharmaceuticals. 00:50:40.630 --> 00:50:47.860 align:middle line:84% And they have thousands of pieces 00:50:47.860 --> 00:50:51.040 align:middle line:84% of dirt, leaf that has been collected around the world 00:50:51.040 --> 00:50:51.910 align:middle line:90% that they test-- 00:50:51.910 --> 00:50:54.340 align:middle line:84% literally some leaf in the rainforest, a piece 00:50:54.340 --> 00:50:55.270 align:middle line:90% of dust in China-- 00:50:55.270 --> 00:50:57.820 align:middle line:84% that they test to see how effective 00:50:57.820 --> 00:51:02.750 align:middle line:84% that it is in inhibiting these 60 different tumors. 00:51:02.750 --> 00:51:05.150 align:middle line:84% And so what we did back then in collaboration 00:51:05.150 --> 00:51:07.400 align:middle line:84% with one of my former interns, Todd Golub, 00:51:07.400 --> 00:51:11.280 align:middle line:84% we extracted the RNA from these 66 cancer cell lines 00:51:11.280 --> 00:51:15.800 align:middle line:84% and measured how much of the RNA was being expressed 00:51:15.800 --> 00:51:19.350 align:middle line:90% in these 60 cancer cell lines. 00:51:19.350 --> 00:51:23.840 align:middle line:84% And we obtained from the National Cancer Institute, 00:51:23.840 --> 00:51:26.150 align:middle line:84% for a subset of the 50,000 compounds 00:51:26.150 --> 00:51:30.440 align:middle line:90% they had tested, 5,000-- 00:51:30.440 --> 00:51:32.600 align:middle line:84% let me see how I'm doing for a time-- 00:51:32.600 --> 00:51:38.330 align:middle line:84% 5,000 drugs that had been tested against the 60 cancer cell 00:51:38.330 --> 00:51:41.660 align:middle line:84% lines and how much they had inhibited 00:51:41.660 --> 00:51:42.980 align:middle line:90% the growth of the cells. 00:51:42.980 --> 00:51:46.220 align:middle line:84% I'll get to the details of this experiment in-- the common cell 00:51:46.220 --> 00:51:48.145 align:middle line:90% details in another lecture. 00:51:48.145 --> 00:51:50.270 align:middle line:84% But suffice it to say, this is the overall picture. 00:51:50.270 --> 00:51:54.350 align:middle line:84% You have 60 common cell lines, 6,000 genes, 00:51:54.350 --> 00:51:56.300 align:middle line:90% 5,000 anticancer agents. 00:51:56.300 --> 00:52:00.800 align:middle line:84% So you have 11,000 variables with only 60 cell lines. 00:52:00.800 --> 00:52:02.720 align:middle line:84% That sounds like a miserable thing to do. 00:52:02.720 --> 00:52:05.330 align:middle line:84% And nonetheless, what we were able to do is the following. 00:52:05.330 --> 00:52:08.210 align:middle line:84% In this complicated looking diagram, 00:52:08.210 --> 00:52:11.090 align:middle line:84% we picked a correlation coefficient between all-- 00:52:11.090 --> 00:52:13.100 align:middle line:84% if we computed all the possible correlations 00:52:13.100 --> 00:52:15.710 align:middle line:84% between all the 11,000 variables, 00:52:15.710 --> 00:52:19.560 align:middle line:84% that gave us on the order of 68 million different correlations. 00:52:19.560 --> 00:52:22.550 align:middle line:84% And if you pick the correlation coefficient threshold of 0.8, 00:52:22.550 --> 00:52:28.890 align:middle line:84% we found 202 networks that were joining 834 variables-- 00:52:28.890 --> 00:52:34.670 align:middle line:84% in other words, only 7% out of the 11,000-- and only 1,200 00:52:34.670 --> 00:52:36.620 align:middle line:90% links out of the 68 million. 00:52:36.620 --> 00:52:39.740 align:middle line:84% 1,200 correlation coefficients out of the 60 million 00:52:39.740 --> 00:52:43.160 align:middle line:90% were above the threshold. 00:52:43.160 --> 00:52:48.320 align:middle line:84% Only one link was above the threshold between a gene 00:52:48.320 --> 00:52:50.960 align:middle line:90% and anticancer agent. 00:52:50.960 --> 00:52:53.900 align:middle line:84% And let me just ask you a quick question. 00:52:53.900 --> 00:52:58.730 align:middle line:84% Why is it that we found a lot of correlations between drug 00:52:58.730 --> 00:53:01.670 align:middle line:84% effect on the cancer cell lines and another drug effect 00:53:01.670 --> 00:53:02.610 align:middle line:90% on cancer cell lines? 00:53:02.610 --> 00:53:06.350 align:middle line:84% So why did we find a lot of high drug-drug correlations, 00:53:06.350 --> 00:53:09.080 align:middle line:84% a lot of high correlations between the inhibitory effect 00:53:09.080 --> 00:53:11.960 align:middle line:84% of one drug and the inhibitory effect of another drug? 00:53:11.960 --> 00:53:13.610 align:middle line:84% Why did we find a lot of correlations, 00:53:13.610 --> 00:53:14.360 align:middle line:90% high correlations? 00:53:14.360 --> 00:53:15.560 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:53:15.560 --> 00:53:15.890 align:middle line:90% ISAAC SAMUEL KOHANE: What? 00:53:15.890 --> 00:53:17.640 align:middle line:84% AUDIENCE: Were they chemically similar? 00:53:17.640 --> 00:53:19.682 align:middle line:84% ISAAC SAMUEL KOHANE: They were chemically similar 00:53:19.682 --> 00:53:22.490 align:middle line:84% because in way a lot of drugs are created 00:53:22.490 --> 00:53:26.817 align:middle line:84% are by copycat drug discovery, doing small permutations 00:53:26.817 --> 00:53:27.650 align:middle line:90% of an existing drug. 00:53:27.650 --> 00:53:29.400 align:middle line:84% So of course, they have similar reactions. 00:53:29.400 --> 00:53:32.180 align:middle line:84% Why were there a lot of gene-gene correlations? 00:53:32.180 --> 00:53:36.570 align:middle line:90% 00:53:36.570 --> 00:53:37.502 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:53:37.502 --> 00:53:38.585 align:middle line:90% ISAAC SAMUEL KOHANE: What? 00:53:38.585 --> 00:53:40.048 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:53:40.048 --> 00:53:41.340 align:middle line:90% ISAAC SAMUEL KOHANE: Not quite. 00:53:41.340 --> 00:53:43.007 align:middle line:84% It's more fundamental insight than that. 00:53:43.007 --> 00:53:44.043 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:53:44.043 --> 00:53:45.960 align:middle line:84% ISAAC SAMUEL KOHANE: Right, because basically, 00:53:45.960 --> 00:53:48.180 align:middle line:84% in order for the darn cell to live, 00:53:48.180 --> 00:53:49.890 align:middle line:84% genes have to have coordinated action. 00:53:49.890 --> 00:53:51.600 align:middle line:84% So not all genes, but the genes that 00:53:51.600 --> 00:53:54.180 align:middle line:84% make the ribosome, that makes the transcription complex that 00:53:54.180 --> 00:53:59.083 align:middle line:84% run oxidation, all have to be synthesized into RNA 00:53:59.083 --> 00:54:00.000 align:middle line:90% at approximately same. 00:54:00.000 --> 00:54:00.760 align:middle line:90% Time. 00:54:00.760 --> 00:54:03.160 align:middle line:84% Otherwise, you're not going to have coordinated action. 00:54:03.160 --> 00:54:05.020 align:middle line:84% There's a lot of exceptions to that. 00:54:05.020 --> 00:54:07.350 align:middle line:90% But by and large, it's true. 00:54:07.350 --> 00:54:10.763 align:middle line:84% And in fact, the whole area of expression analysis clustering, 00:54:10.763 --> 00:54:12.180 align:middle line:84% as I'll get to in another lecture, 00:54:12.180 --> 00:54:15.120 align:middle line:84% would not work if that intuition was not true. 00:54:15.120 --> 00:54:17.790 align:middle line:84% But God or evolution did not evolve 00:54:17.790 --> 00:54:21.210 align:middle line:84% genes and chemotherapeutic compounds together. 00:54:21.210 --> 00:54:22.890 align:middle line:84% So there's no particular prior reason 00:54:22.890 --> 00:54:26.910 align:middle line:84% why the inhibitory pattern of a drug and the gene expression 00:54:26.910 --> 00:54:30.390 align:middle line:90% profile should be linked. 00:54:30.390 --> 00:54:34.140 align:middle line:84% And so we were darn lucky to find even one, but we did. 00:54:34.140 --> 00:54:43.560 align:middle line:84% And it turned out that links were this gene and this drug. 00:54:43.560 --> 00:54:46.140 align:middle line:84% Before I explain to you what they are, 00:54:46.140 --> 00:54:48.090 align:middle line:90% let me summarize what it means. 00:54:48.090 --> 00:54:50.130 align:middle line:84% It means that across all these 60 cancer cell 00:54:50.130 --> 00:54:52.170 align:middle line:84% lines-- and these were different very different cancers cell 00:54:52.170 --> 00:54:52.670 align:middle line:90% lines. 00:54:52.670 --> 00:54:57.630 align:middle line:84% They were liver cancer, breast cancer, skin cancer, blood 00:54:57.630 --> 00:55:01.080 align:middle line:90% cancers, 60 different cancers. 00:55:01.080 --> 00:55:04.500 align:middle line:84% The more of this gene was expression in the cancer, 00:55:04.500 --> 00:55:07.680 align:middle line:84% the more sensitive the cancer cell line 00:55:07.680 --> 00:55:10.920 align:middle line:84% was to this chemotherapeutic agent. 00:55:10.920 --> 00:55:15.060 align:middle line:84% And it turns out that this gene is L-plastin, 00:55:15.060 --> 00:55:19.530 align:middle line:84% also known as LCP1 or Lymphocyte Cytosolic Protein. 00:55:19.530 --> 00:55:21.510 align:middle line:84% The drug is this unpronounceable thing. 00:55:21.510 --> 00:55:23.190 align:middle line:84% I mean, a lot of these are just drugs 00:55:23.190 --> 00:55:24.900 align:middle line:84% that they obtained essentially wholesale 00:55:24.900 --> 00:55:26.220 align:middle line:90% from pharmaceutical companies. 00:55:26.220 --> 00:55:28.710 align:middle line:84% But it turns out it's a [INAUDIBLE] 00:55:28.710 --> 00:55:30.525 align:middle line:90% carboxylic acid derivative. 00:55:30.525 --> 00:55:32.400 align:middle line:84% Those of you who know anything about diabetes 00:55:32.400 --> 00:55:35.910 align:middle line:84% know that other [INAUDIBLE] carboxylic acid derivatives 00:55:35.910 --> 00:55:37.960 align:middle line:84% are known to inhibit tumor cell growth. 00:55:37.960 --> 00:55:39.690 align:middle line:84% And by the way, these are the same class 00:55:39.690 --> 00:55:42.480 align:middle line:84% of drugs as the oral hypoglycemic agents, 00:55:42.480 --> 00:55:45.750 align:middle line:84% the pills that type two diabetics will take. 00:55:45.750 --> 00:55:48.070 align:middle line:84% And our role for this gene in [? tumorigenicity, ?] 00:55:48.070 --> 00:55:51.300 align:middle line:84% the generation of tumors, have previously been postulated. 00:55:51.300 --> 00:55:53.070 align:middle line:84% And in fact, subsequent to our publication 00:55:53.070 --> 00:55:55.390 align:middle line:84% of this paper in [INAUDIBLE],, in fact, 00:55:55.390 --> 00:55:57.540 align:middle line:84% a group at the National Cancer Institute 00:55:57.540 --> 00:56:01.260 align:middle line:84% was able to show the dose response that we had, in fact, 00:56:01.260 --> 00:56:01.770 align:middle line:90% calculated. 00:56:01.770 --> 00:56:02.820 align:middle line:90% Yes. 00:56:02.820 --> 00:56:05.980 align:middle line:84% AUDIENCE: So if I'm understanding the description 00:56:05.980 --> 00:56:08.003 align:middle line:84% of the [INAUDIBLE] experiment and the results-- 00:56:08.003 --> 00:56:09.045 align:middle line:90% ISAAC SAMUEL KOHANE: Yes. 00:56:09.045 --> 00:56:12.660 align:middle line:90% 00:56:12.660 --> 00:56:15.160 align:middle line:84% AUDIENCE: --is it safe to say as a generalization 00:56:15.160 --> 00:56:19.940 align:middle line:84% that whenever you find a link between a particular 00:56:19.940 --> 00:56:24.780 align:middle line:84% [INAUDIBLE] and a gene that [INAUDIBLE] 00:56:24.780 --> 00:56:27.900 align:middle line:84% across the [INAUDIBLE] cell lines 00:56:27.900 --> 00:56:30.180 align:middle line:84% these are genes that are typically 00:56:30.180 --> 00:56:32.580 align:middle line:84% going to have a high conservation 00:56:32.580 --> 00:56:33.830 align:middle line:90% rate across the [INAUDIBLE]? 00:56:33.830 --> 00:56:34.830 align:middle line:90% ISAAC SAMUEL KOHANE: No. 00:56:34.830 --> 00:56:35.705 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:56:35.705 --> 00:56:38.490 align:middle line:84% ISAAC SAMUEL KOHANE: In fact, if it had been true 00:56:38.490 --> 00:56:40.410 align:middle line:84% that the gene had a high concentration 00:56:40.410 --> 00:56:42.060 align:middle line:84% rate across all 60 cancer cell lines-- 00:56:42.060 --> 00:56:43.480 align:middle line:90% AUDIENCE: [INAUDIBLE]. 00:56:43.480 --> 00:56:47.670 align:middle line:84% In other words, like, this gene data [INAUDIBLE]---- 00:56:47.670 --> 00:56:48.900 align:middle line:90% ISAAC SAMUEL KOHANE: Yes. 00:56:48.900 --> 00:56:53.670 align:middle line:84% AUDIENCE: --that this drug is acting on the different cell 00:56:53.670 --> 00:56:56.165 align:middle line:84% lines, is acting on this one particular gene. 00:56:56.165 --> 00:56:58.290 align:middle line:84% But there probably are differences in [INAUDIBLE].. 00:56:58.290 --> 00:56:58.950 align:middle line:84% ISAAC SAMUEL KOHANE: Let's just have 00:56:58.950 --> 00:57:00.630 align:middle line:84% our conversation [INAUDIBLE] for the rest of the class. 00:57:00.630 --> 00:57:01.830 align:middle line:90% We don't know how it's acting. 00:57:01.830 --> 00:57:03.788 align:middle line:84% All we're observing right now is a correlation. 00:57:03.788 --> 00:57:05.040 align:middle line:90% AUDIENCE: OK, right. 00:57:05.040 --> 00:57:06.395 align:middle line:90% ISAAC SAMUEL KOHANE: Right? 00:57:06.395 --> 00:57:08.520 align:middle line:84% But I still think that you have a question in there 00:57:08.520 --> 00:57:12.240 align:middle line:84% which is still true regardless of what I just said. 00:57:12.240 --> 00:57:15.870 align:middle line:84% So the more of this gene expressed, the cancer cell line 00:57:15.870 --> 00:57:18.600 align:middle line:84% is going to be provably more sensitive 00:57:18.600 --> 00:57:19.923 align:middle line:90% to the chemotherapeutic agents. 00:57:19.923 --> 00:57:21.090 align:middle line:90% So you had a question still. 00:57:21.090 --> 00:57:21.300 align:middle line:90% AUDIENCE: Right. 00:57:21.300 --> 00:57:23.070 align:middle line:84% Well, it was that it didn't have to do 00:57:23.070 --> 00:57:30.058 align:middle line:84% with the concentration of the gene but rather just that the-- 00:57:30.058 --> 00:57:32.100 align:middle line:84% just for argument's sake, let's say it's actually 00:57:32.100 --> 00:57:33.017 align:middle line:90% acting on [INAUDIBLE]. 00:57:33.017 --> 00:57:34.267 align:middle line:90% ISAAC SAMUEL KOHANE: OK, fine. 00:57:34.267 --> 00:57:34.880 align:middle line:90% Yeah. 00:57:34.880 --> 00:57:35.380 align:middle line:90% Right. 00:57:35.380 --> 00:57:37.190 align:middle line:90% AUDIENCE: [INAUDIBLE]. 00:57:37.190 --> 00:57:39.600 align:middle line:84% So let's say it's acting on the gene product 00:57:39.600 --> 00:57:43.770 align:middle line:84% and across the cell lines, then it would be a safe assumption 00:57:43.770 --> 00:57:46.240 align:middle line:84% that the gene product would be very similar. 00:57:46.240 --> 00:57:49.260 align:middle line:84% In other words, there would be this [INAUDIBLE].. 00:57:49.260 --> 00:57:52.590 align:middle line:84% Well, there's not-- whereas with other gene products, 00:57:52.590 --> 00:57:55.857 align:middle line:84% where there is no correlation between a therapeutic agent-- 00:57:55.857 --> 00:57:56.940 align:middle line:90% ISAAC SAMUEL KOHANE: Yeah. 00:57:56.940 --> 00:58:00.150 align:middle line:84% AUDIENCE: --you have a high variability of different types 00:58:00.150 --> 00:58:01.248 align:middle line:90% within those cells. 00:58:01.248 --> 00:58:03.790 align:middle line:84% ISAAC SAMUEL KOHANE: I may be misunderstanding your question, 00:58:03.790 --> 00:58:05.332 align:middle line:84% and I'll give you back exactly what I 00:58:05.332 --> 00:58:06.960 align:middle line:90% believe my misunderstanding is. 00:58:06.960 --> 00:58:10.140 align:middle line:84% And maybe it's yours, and we'll clarify it right now. 00:58:10.140 --> 00:58:11.910 align:middle line:84% For a given gene, what we're measuring 00:58:11.910 --> 00:58:16.200 align:middle line:84% is a gene product, which is the amount of RNA being expressed. 00:58:16.200 --> 00:58:20.970 align:middle line:84% And basically, what we know is that the gene 00:58:20.970 --> 00:58:24.490 align:middle line:84% product is the same for every cancer cell line. 00:58:24.490 --> 00:58:27.750 align:middle line:84% It's just they don't have a new gene. 00:58:27.750 --> 00:58:29.820 align:middle line:84% They have the same complement of genes. 00:58:29.820 --> 00:58:32.432 align:middle line:84% But just from the point of view point of view 00:58:32.432 --> 00:58:34.140 align:middle line:84% of this experiment, the only thing that's 00:58:34.140 --> 00:58:35.640 align:middle line:84% different between the 60 cancer cell 00:58:35.640 --> 00:58:37.680 align:middle line:84% lines in a very reductionist point of view, 00:58:37.680 --> 00:58:39.180 align:middle line:84% just looking at the gene expression, 00:58:39.180 --> 00:58:42.720 align:middle line:84% is how much of each of those genes they're making. 00:58:42.720 --> 00:58:44.790 align:middle line:84% AUDIENCE: So there aren't any [INAUDIBLE].. 00:58:44.790 --> 00:58:47.520 align:middle line:84% ISAAC SAMUEL KOHANE: No, you're addressing 00:58:47.520 --> 00:58:50.130 align:middle line:84% a whole other layer of complexity 00:58:50.130 --> 00:58:51.660 align:middle line:90% that may be very, very true. 00:58:51.660 --> 00:58:52.785 align:middle line:90% But we just didn't measure. 00:58:52.785 --> 00:58:54.480 align:middle line:84% AUDIENCE: All right, because I thought 00:58:54.480 --> 00:58:56.760 align:middle line:84% that one of the take home messages 00:58:56.760 --> 00:59:00.867 align:middle line:84% was that for sort of fundamental gene products within a pathway. 00:59:00.867 --> 00:59:01.950 align:middle line:90% ISAAC SAMUEL KOHANE: Yeah. 00:59:01.950 --> 00:59:03.492 align:middle line:84% AUDIENCE: Those are what you're going 00:59:03.492 --> 00:59:07.830 align:middle line:84% to find links to therapeutic agents more than, 00:59:07.830 --> 00:59:10.353 align:middle line:90% say, something, you know-- 00:59:10.353 --> 00:59:12.270 align:middle line:84% ISAAC SAMUEL KOHANE: Well, we got out of this. 00:59:12.270 --> 00:59:15.840 align:middle line:84% So the answer is not in the way that you conceive it 00:59:15.840 --> 00:59:16.925 align:middle line:90% because if you're doing-- 00:59:16.925 --> 00:59:18.300 align:middle line:84% and by the way, it's a good point 00:59:18.300 --> 00:59:19.890 align:middle line:84% because it reminds us again of the distinction 00:59:19.890 --> 00:59:22.590 align:middle line:84% I drew at the beginning where that's a very simple diagram 00:59:22.590 --> 00:59:25.163 align:middle line:90% about DNA, RNA, and protein. 00:59:25.163 --> 00:59:26.580 align:middle line:84% There would be a study where you'd 00:59:26.580 --> 00:59:29.610 align:middle line:84% be looking at the genotype and saying-- 00:59:29.610 --> 00:59:31.770 align:middle line:84% doing the very same study and looking 00:59:31.770 --> 00:59:34.110 align:middle line:84% at all the possible genotypes and saying, 00:59:34.110 --> 00:59:36.120 align:middle line:84% which genotypes are the most predictive 00:59:36.120 --> 00:59:39.580 align:middle line:90% of this kind of expression? 00:59:39.580 --> 00:59:43.020 align:middle line:84% But here, we're not worrying about individual differences 00:59:43.020 --> 00:59:44.910 align:middle line:84% in the genes we're just looking at the amount 00:59:44.910 --> 00:59:46.230 align:middle line:90% of the gene being expressed. 00:59:46.230 --> 00:59:52.210 align:middle line:84% Nonetheless, if there was, for instance, a direct effect-- 00:59:52.210 --> 00:59:54.570 align:middle line:84% in other words, if the more this gene is expressed, 00:59:54.570 --> 00:59:57.000 align:middle line:84% the more sensitive the cancer is, short of actually 00:59:57.000 --> 00:59:58.500 align:middle line:84% going and understanding the biology, 00:59:58.500 --> 01:00:01.390 align:middle line:84% which is the right answer, you still ask yourself a question. 01:00:01.390 --> 01:00:04.980 align:middle line:84% So if I make this gene more highly expressed with a vector 01:00:04.980 --> 01:00:07.290 align:middle line:84% to make it more expressed, will this cell line 01:00:07.290 --> 01:00:11.130 align:middle line:84% become more resistant, more sensitive to chemotherapy? 01:00:11.130 --> 01:00:14.610 align:middle line:84% That becomes a therapy in and of itself. 01:00:14.610 --> 01:00:16.330 align:middle line:90% But thanks for asking question. 01:00:16.330 --> 01:00:18.390 align:middle line:84% So just now-- and this would give you 01:00:18.390 --> 01:00:22.020 align:middle line:84% an idea why in The Economist just appeared within last two 01:00:22.020 --> 01:00:24.780 align:middle line:84% weeks the following cartoon that was part of an article 01:00:24.780 --> 01:00:26.820 align:middle line:84% called "Malignant [? Mass," ?] where they're 01:00:26.820 --> 01:00:30.270 align:middle line:84% arguing that actually, coming up with these comprehensive models 01:00:30.270 --> 01:00:34.050 align:middle line:84% of which this is a very simple example of susceptibility 01:00:34.050 --> 01:00:36.630 align:middle line:84% of cancers to treatment and risk to cancers 01:00:36.630 --> 01:00:38.670 align:middle line:84% is actually become so mathematical that we can 01:00:38.670 --> 01:00:42.695 align:middle line:84% actually have a whole new class of researchers who are going 01:00:42.695 --> 01:00:44.070 align:middle line:84% to take advantage of these models 01:00:44.070 --> 01:00:46.620 align:middle line:84% to actually find new therapeutic agents. 01:00:46.620 --> 01:00:49.190 align:middle line:90% 01:00:49.190 --> 01:00:50.080 align:middle line:90% So let's go to-- 01:00:50.080 --> 01:00:53.410 align:middle line:84% we've talked about how much of these data 01:00:53.410 --> 01:00:55.300 align:middle line:84% measurement techniques are available today 01:00:55.300 --> 01:00:56.480 align:middle line:90% in a large scale. 01:00:56.480 --> 01:00:58.570 align:middle line:84% We've discussed [INAUDIBLE] genetic. 01:00:58.570 --> 01:01:00.520 align:middle line:84% Let's now talk about hereditary-- 01:01:00.520 --> 01:01:01.690 align:middle line:90% heredity, rather. 01:01:01.690 --> 01:01:03.730 align:middle line:84% So the way a population geneticist 01:01:03.730 --> 01:01:06.490 align:middle line:84% thinks about heredity in the so-called broad sense 01:01:06.490 --> 01:01:09.370 align:middle line:84% with a capital H is the following. 01:01:09.370 --> 01:01:12.280 align:middle line:84% Heredity includes all genetic influences 01:01:12.280 --> 01:01:15.010 align:middle line:84% on the phenotype whether due to additive, dominant, 01:01:15.010 --> 01:01:16.430 align:middle line:90% or interactive effects. 01:01:16.430 --> 01:01:20.830 align:middle line:84% So when you hear in a technical paper heredity, what they mean 01:01:20.830 --> 01:01:24.100 align:middle line:84% is the ratio of the variance due to the genotype 01:01:24.100 --> 01:01:26.890 align:middle line:84% over the total variance of the phenotype where 01:01:26.890 --> 01:01:32.410 align:middle line:84% the variance of the genotype is additive variance, 01:01:32.410 --> 01:01:36.430 align:middle line:84% dominant variance, and interactive variance 01:01:36.430 --> 01:01:37.540 align:middle line:90% between the genes. 01:01:37.540 --> 01:01:42.920 align:middle line:84% That's the technical definition of heritability. 01:01:42.920 --> 01:01:46.000 align:middle line:84% So let's think about that in a concrete fashion. 01:01:46.000 --> 01:01:48.580 align:middle line:84% Don't some people just eat and not get fat? 01:01:48.580 --> 01:01:49.990 align:middle line:90% Isn't that true? 01:01:49.990 --> 01:01:51.490 align:middle line:90% Isn't it in their genes? 01:01:51.490 --> 01:01:53.800 align:middle line:84% In other words, these poor people who 01:01:53.800 --> 01:01:55.300 align:middle line:84% just, no matter how little they eat, 01:01:55.300 --> 01:01:58.030 align:middle line:84% they're just going to balloon up? 01:01:58.030 --> 01:01:59.530 align:middle line:90% It's just tough. 01:01:59.530 --> 01:02:08.350 align:middle line:84% And it's really their problem that they 01:02:08.350 --> 01:02:10.510 align:middle line:84% have this genetic background that makes them 01:02:10.510 --> 01:02:14.050 align:middle line:90% at risk for this obesity. 01:02:14.050 --> 01:02:17.675 align:middle line:90% So we've got a real big problem. 01:02:17.675 --> 01:02:19.550 align:middle line:84% So the way to think about it is an experiment 01:02:19.550 --> 01:02:23.900 align:middle line:84% that I only dared to do four years ago at MIT. 01:02:23.900 --> 01:02:28.460 align:middle line:84% I asked all the Asian students in the classroom 01:02:28.460 --> 01:02:29.600 align:middle line:90% to raise their hand. 01:02:29.600 --> 01:02:33.800 align:middle line:84% They raised their hand and I told them to leave. 01:02:33.800 --> 01:02:36.710 align:middle line:84% They're upset with me, and they walk out of the room. 01:02:36.710 --> 01:02:39.840 align:middle line:84% Then of the remaining students in the class, 01:02:39.840 --> 01:02:43.100 align:middle line:84% which, since this was MIT, was about 30% left, 01:02:43.100 --> 01:02:47.000 align:middle line:90% I say, take your pulse. 01:02:47.000 --> 01:02:49.370 align:middle line:84% And we took a pulse, and I found a nerd in the room. 01:02:49.370 --> 01:02:53.180 align:middle line:84% And I said, fire up Excel, and everybody called out 01:02:53.180 --> 01:02:55.250 align:middle line:90% their pulses. 01:02:55.250 --> 01:02:57.290 align:middle line:84% And I had the nerd take the average 01:02:57.290 --> 01:02:58.700 align:middle line:90% and the standard deviation. 01:02:58.700 --> 01:03:01.430 align:middle line:84% Welcomed back the Asian students and went 01:03:01.430 --> 01:03:03.890 align:middle line:84% through the same process with them. 01:03:03.890 --> 01:03:05.390 align:middle line:84% Everybody takes their heart rate. 01:03:05.390 --> 01:03:09.170 align:middle line:84% We tally it, mean, standard deviation. 01:03:09.170 --> 01:03:12.290 align:middle line:90% And then we show them the board. 01:03:12.290 --> 01:03:15.440 align:middle line:84% And in fact, there's a more than two 01:03:15.440 --> 01:03:18.518 align:middle line:84% standard deviation separation between the Asians 01:03:18.518 --> 01:03:19.310 align:middle line:90% and the non-Asians. 01:03:19.310 --> 01:03:21.980 align:middle line:84% And I explain to them, as you well know, 01:03:21.980 --> 01:03:24.950 align:middle line:84% the basal heart rate of non-Asians 01:03:24.950 --> 01:03:29.090 align:middle line:84% is quite a bit lower than that of Asians. 01:03:29.090 --> 01:03:30.463 align:middle line:90% And they sort of look puzzled. 01:03:30.463 --> 01:03:31.130 align:middle line:90% Did I know that? 01:03:31.130 --> 01:03:31.680 align:middle line:90% Really? 01:03:31.680 --> 01:03:32.760 align:middle line:90% Really? 01:03:32.760 --> 01:03:33.260 align:middle line:90% Maybe. 01:03:33.260 --> 01:03:33.980 align:middle line:90% I don't know. 01:03:33.980 --> 01:03:36.770 align:middle line:84% And then I say, well, is that really right? 01:03:36.770 --> 01:03:40.260 align:middle line:84% And then suddenly it dawns on what's happened here. 01:03:40.260 --> 01:03:41.510 align:middle line:90% We're not talking about genes. 01:03:41.510 --> 01:03:43.590 align:middle line:90% What are we talking about? 01:03:43.590 --> 01:03:44.090 align:middle line:90% What? 01:03:44.090 --> 01:03:45.082 align:middle line:90% AUDIENCE: Work. 01:03:45.082 --> 01:03:47.540 align:middle line:84% ISAAC SAMUEL KOHANE: Work-- we're talking about environment 01:03:47.540 --> 01:03:50.120 align:middle line:84% because basically, I pissed them off by doing 01:03:50.120 --> 01:03:52.280 align:middle line:90% a potentially racist thing. 01:03:52.280 --> 01:03:54.830 align:middle line:84% So the catecholamines were pouring through their veins, 01:03:54.830 --> 01:03:56.930 align:middle line:90% and I made them get up. 01:03:56.930 --> 01:04:00.680 align:middle line:84% And so I did two major environmental stimuli 01:04:00.680 --> 01:04:04.220 align:middle line:90% which changed their physiology. 01:04:04.220 --> 01:04:06.380 align:middle line:84% But if you didn't know that, that hidden variable, 01:04:06.380 --> 01:04:08.265 align:middle line:84% you could legitimately come to a conclusion 01:04:08.265 --> 01:04:09.890 align:middle line:84% that there's some racial stratification 01:04:09.890 --> 01:04:11.348 align:middle line:84% around this phenotype and therefore 01:04:11.348 --> 01:04:13.010 align:middle line:90% is with reduction to genes. 01:04:13.010 --> 01:04:17.510 align:middle line:84% And you could ask the same thing about the overweight people. 01:04:17.510 --> 01:04:21.650 align:middle line:84% You can say, look, they're eating just as much as I am, 01:04:21.650 --> 01:04:26.340 align:middle line:84% and somehow, they're getting fat and I'm not. 01:04:26.340 --> 01:04:29.670 align:middle line:84% So obesity is just a massive epidemic, and that's the point. 01:04:29.670 --> 01:04:31.110 align:middle line:90% It's an epidemic. 01:04:31.110 --> 01:04:33.510 align:middle line:84% This is the prevalence of obesity, note, 01:04:33.510 --> 01:04:36.870 align:middle line:90% just from in the last 40 years. 01:04:36.870 --> 01:04:38.400 align:middle line:90% Exponential rise. 01:04:38.400 --> 01:04:42.720 align:middle line:84% Now I defy you to find any possible genetic evolutionary 01:04:42.720 --> 01:04:45.810 align:middle line:84% evolutionary model that explains how our genome changed 01:04:45.810 --> 01:04:47.400 align:middle line:90% so fast in 30 years. 01:04:47.400 --> 01:04:49.553 align:middle line:84% And note, by the way, this was the same time 01:04:49.553 --> 01:04:50.970 align:middle line:84% as the amount of calories from fat 01:04:50.970 --> 01:04:53.460 align:middle line:90% has been steadily decreasing. 01:04:53.460 --> 01:04:55.300 align:middle line:84% In fact, some argue that's the cause of it, 01:04:55.300 --> 01:04:56.842 align:middle line:84% by the way, because we've substituted 01:04:56.842 --> 01:04:59.310 align:middle line:84% that those fat calories with carbohydrate calories. 01:04:59.310 --> 01:05:00.693 align:middle line:90% AUDIENCE: But it's US only. 01:05:00.693 --> 01:05:02.110 align:middle line:84% ISAAC SAMUEL KOHANE: It's US only. 01:05:02.110 --> 01:05:04.493 align:middle line:90% By the way-- but there's a lag. 01:05:04.493 --> 01:05:05.910 align:middle line:84% Even third world countries are now 01:05:05.910 --> 01:05:08.400 align:middle line:84% beginning-- the moment they become caloric sufficient, 01:05:08.400 --> 01:05:11.830 align:middle line:90% they actually are showing this. 01:05:11.830 --> 01:05:15.550 align:middle line:84% And so obviously, it's not genetic, 01:05:15.550 --> 01:05:18.270 align:middle line:84% and it's actually conditioned on some significant change 01:05:18.270 --> 01:05:20.760 align:middle line:84% in our environment most likely having 01:05:20.760 --> 01:05:23.550 align:middle line:84% to do with our combination of our exercise patterns 01:05:23.550 --> 01:05:27.180 align:middle line:90% and our dietary composition. 01:05:27.180 --> 01:05:30.420 align:middle line:84% And yet if you did not understand that, 01:05:30.420 --> 01:05:32.970 align:middle line:84% you'd just be hammering on the genome. 01:05:32.970 --> 01:05:35.110 align:middle line:84% And it's that kind of reductionism, 01:05:35.110 --> 01:05:37.200 align:middle line:90% that we have to avoid. 01:05:37.200 --> 01:05:40.050 align:middle line:84% But if you-- in terms of genetic medicine, not genomic medicine, 01:05:40.050 --> 01:05:41.730 align:middle line:84% where you really think comprehensively, 01:05:41.730 --> 01:05:44.590 align:middle line:84% you could easily run into this kind of error. 01:05:44.590 --> 01:05:46.440 align:middle line:84% And so how do we define environment, diet, 01:05:46.440 --> 01:05:49.320 align:middle line:84% daily habits, environmental insults, medical care? 01:05:49.320 --> 01:05:53.620 align:middle line:84% And so the genotype does not capture the individual patient 01:05:53.620 --> 01:05:54.120 align:middle line:90% states. 01:05:54.120 --> 01:05:55.537 align:middle line:84% It's what I told you before, but I 01:05:55.537 --> 01:05:57.623 align:middle line:84% think it's a graphical example of how that's true. 01:05:57.623 --> 01:05:59.040 align:middle line:84% So we need to capture and quantify 01:05:59.040 --> 01:06:00.255 align:middle line:90% the environmental influences. 01:06:00.255 --> 01:06:02.130 align:middle line:84% We need to capture the effect of the genotype 01:06:02.130 --> 01:06:04.800 align:middle line:84% and the environmental effects on the phenotype. 01:06:04.800 --> 01:06:08.130 align:middle line:84% And these two comprise-- history, physical, 01:06:08.130 --> 01:06:11.017 align:middle line:84% laboratory studies, imaging, which, in fact, you'll 01:06:11.017 --> 01:06:13.350 align:middle line:84% all recognize those of you getting your medical training 01:06:13.350 --> 01:06:16.420 align:middle line:90% is the basic medical history. 01:06:16.420 --> 01:06:22.140 align:middle line:84% And so if we really want to do the comprehensive kind 01:06:22.140 --> 01:06:24.090 align:middle line:84% of studies that we need to be able to dissect 01:06:24.090 --> 01:06:25.920 align:middle line:84% environmental and genetic interactions, 01:06:25.920 --> 01:06:27.690 align:middle line:84% we have to have these data items. 01:06:27.690 --> 01:06:29.873 align:middle line:84% And that's why this is part of genomic medicine-- 01:06:29.873 --> 01:06:31.290 align:middle line:84% because those of you who are going 01:06:31.290 --> 01:06:32.832 align:middle line:84% to be doctors or clinical researchers 01:06:32.832 --> 01:06:36.245 align:middle line:84% or researchers studying clinical phenomenon, 01:06:36.245 --> 01:06:38.370 align:middle line:84% somehow you're going to have to be able to get this 01:06:38.370 --> 01:06:40.410 align:middle line:90% accurately and at large scale. 01:06:40.410 --> 01:06:44.730 align:middle line:84% And that's kind of boring, frankly, but nontrivial. 01:06:44.730 --> 01:06:48.365 align:middle line:90% 01:06:48.365 --> 01:06:50.240 align:middle line:84% And so this is just repeating what I've said, 01:06:50.240 --> 01:06:51.590 align:middle line:84% but I just really want to emphasize it. 01:06:51.590 --> 01:06:54.050 align:middle line:84% There's more to the state description than the genome. 01:06:54.050 --> 01:06:56.900 align:middle line:84% Given the necessity to capture both environment genetic state 01:06:56.900 --> 01:06:58.820 align:middle line:84% and the interaction, it's only then 01:06:58.820 --> 01:07:03.260 align:middle line:84% that we'll be able to elucidate the variation of environment 01:07:03.260 --> 01:07:05.330 align:middle line:84% due to the genome and through the interaction 01:07:05.330 --> 01:07:07.400 align:middle line:84% between the environment and the genome. 01:07:07.400 --> 01:07:11.150 align:middle line:84% For example, you're only going to figure out 01:07:11.150 --> 01:07:16.040 align:middle line:84% the risk effects of smoking on lung cancer 01:07:16.040 --> 01:07:19.310 align:middle line:84% if you can quantify well the environment, which 01:07:19.310 --> 01:07:23.690 align:middle line:84% is both the smoking and the exposure 01:07:23.690 --> 01:07:28.490 align:middle line:84% to other harmful chemicals in the environment. 01:07:28.490 --> 01:07:30.530 align:middle line:84% And it's required for effective new therapies. 01:07:30.530 --> 01:07:33.240 align:middle line:84% It's required for deeper understanding of mechanism, 01:07:33.240 --> 01:07:35.615 align:middle line:84% and it requires capturing the aforementioned interaction, 01:07:35.615 --> 01:07:39.020 align:middle line:84% and the less we capture, the more undetermined, in the sense 01:07:39.020 --> 01:07:41.360 align:middle line:84% that I gave to you, the system is. 01:07:41.360 --> 01:07:45.460 align:middle line:90% 01:07:45.460 --> 01:07:47.500 align:middle line:90% Well, we're running out of time. 01:07:47.500 --> 01:07:49.390 align:middle line:84% One thing I want to make true for this class 01:07:49.390 --> 01:07:51.050 align:middle line:90% is that I always end on time. 01:07:51.050 --> 01:07:54.220 align:middle line:84% So we'll just talk about the last bullet, 01:07:54.220 --> 01:07:57.130 align:middle line:84% and at another time, I'll talk to you about accelerating 01:07:57.130 --> 01:07:59.740 align:middle line:90% consumer activation. 01:07:59.740 --> 01:08:02.920 align:middle line:84% Here's a standard pediatric question. 01:08:02.920 --> 01:08:06.220 align:middle line:84% You have here the standard growth curves 01:08:06.220 --> 01:08:08.530 align:middle line:90% in height and in weight. 01:08:08.530 --> 01:08:10.570 align:middle line:84% And you have a patient who's falling off 01:08:10.570 --> 01:08:13.990 align:middle line:84% their height curve and sort of falling off the weight curve. 01:08:13.990 --> 01:08:15.850 align:middle line:84% And in order to diagnose them, you 01:08:15.850 --> 01:08:18.177 align:middle line:84% have heights and weights and family history. 01:08:18.177 --> 01:08:19.719 align:middle line:84% You can take an X-ray of their wrist. 01:08:19.719 --> 01:08:22.810 align:middle line:84% You can measure breast development or size 01:08:22.810 --> 01:08:24.310 align:middle line:90% of the testicles. 01:08:24.310 --> 01:08:27.250 align:middle line:84% And the disorders show characteristic patterns 01:08:27.250 --> 01:08:28.401 align:middle line:90% on a growth chart. 01:08:28.401 --> 01:08:29.859 align:middle line:84% But if you just have the misfortune 01:08:29.859 --> 01:08:32.979 align:middle line:84% of sending them to my clinic, we'll also do some other tests. 01:08:32.979 --> 01:08:35.170 align:middle line:90% We'll look at thyroid function. 01:08:35.170 --> 01:08:39.069 align:middle line:84% We'll look at a protein called IGF1 01:08:39.069 --> 01:08:41.470 align:middle line:84% that's made in response to growth hormone. 01:08:41.470 --> 01:08:43.990 align:middle line:84% We look for inflammation, looking at your blood count, 01:08:43.990 --> 01:08:45.140 align:middle line:90% and so on. 01:08:45.140 --> 01:08:49.270 align:middle line:84% And if we still have evidence that you're really not going 01:08:49.270 --> 01:08:52.660 align:middle line:84% well, then we bring you into the hospital 01:08:52.660 --> 01:08:58.090 align:middle line:84% where we expose you to insulin and to glucagon. 01:08:58.090 --> 01:09:00.550 align:middle line:84% And that makes a kid hypoglycemic. 01:09:00.550 --> 01:09:02.290 align:middle line:84% So they pass out and feel crummy, 01:09:02.290 --> 01:09:05.170 align:middle line:84% and then they get nauseous from the glucagon. 01:09:05.170 --> 01:09:07.149 align:middle line:84% So they puke, and then they pay $2,000 01:09:07.149 --> 01:09:10.899 align:middle line:84% for the pleasure of having been with us in the morning. 01:09:10.899 --> 01:09:14.109 align:middle line:84% And the interpretation remains controversial 01:09:14.109 --> 01:09:16.843 align:middle line:84% nonetheless because there's a significant false positive 01:09:16.843 --> 01:09:19.010 align:middle line:84% rate, and there's a significant false negative rate. 01:09:19.010 --> 01:09:19.677 align:middle line:90% And why is that? 01:09:19.677 --> 01:09:21.093 align:middle line:84% Because we're really not capturing 01:09:21.093 --> 01:09:22.090 align:middle line:90% the underlying process. 01:09:22.090 --> 01:09:24.460 align:middle line:84% And for instance, obese patients sometimes don't quite 01:09:24.460 --> 01:09:26.330 align:middle line:90% secrete enough growth hormone. 01:09:26.330 --> 01:09:27.850 align:middle line:84% If you're before puberty, sometimes 01:09:27.850 --> 01:09:29.725 align:middle line:84% if you don't have enough sex hormones at all, 01:09:29.725 --> 01:09:35.859 align:middle line:84% you might not, in fact, secrete any growth hormone 01:09:35.859 --> 01:09:37.390 align:middle line:84% even though you're ultimately going 01:09:37.390 --> 01:09:39.130 align:middle line:90% to secrete growth hormone. 01:09:39.130 --> 01:09:43.450 align:middle line:84% And we're totally going to miss the following thing. 01:09:43.450 --> 01:09:47.710 align:middle line:84% Here at the tip of the X and Y chromosome 01:09:47.710 --> 01:09:52.689 align:middle line:84% is a gene called SHOX, Short Stature Homeobox. 01:09:52.689 --> 01:09:57.644 align:middle line:84% And those of you who are doctors or medical students, 01:09:57.644 --> 01:09:59.770 align:middle line:84% you know what Turner syndrome is? 01:09:59.770 --> 01:10:02.220 align:middle line:90% 01:10:02.220 --> 01:10:03.720 align:middle line:84% Do you know what Turner syndrome is? 01:10:03.720 --> 01:10:04.500 align:middle line:90% AUDIENCE: [INAUDIBLE] 01:10:04.500 --> 01:10:05.625 align:middle line:90% ISAAC SAMUEL KOHANE: Close. 01:10:05.625 --> 01:10:06.810 align:middle line:90% You have the right category. 01:10:06.810 --> 01:10:10.320 align:middle line:84% X no Y, and they look like girls. 01:10:10.320 --> 01:10:12.660 align:middle line:90% And they're short. 01:10:12.660 --> 01:10:14.790 align:middle line:90% What [? neck? ?] Excellent. 01:10:14.790 --> 01:10:15.570 align:middle line:90% They're short. 01:10:15.570 --> 01:10:23.700 align:middle line:84% They only have an X and no Y. And after class, I'll 01:10:23.700 --> 01:10:27.240 align:middle line:84% give you the full Turner syndrome so I can end on time. 01:10:27.240 --> 01:10:28.098 align:middle line:90% And they're short. 01:10:28.098 --> 01:10:28.890 align:middle line:90% Why are they short? 01:10:28.890 --> 01:10:33.610 align:middle line:84% Because they only have one dose of SHOX. 01:10:33.610 --> 01:10:36.150 align:middle line:90% Now it turns out-- 01:10:36.150 --> 01:10:39.480 align:middle line:84% because they don't have two X chromosomes. 01:10:39.480 --> 01:10:42.180 align:middle line:84% It turns out the most common chromosome is, of course, 01:10:42.180 --> 01:10:44.880 align:middle line:84% of short stature in males, and it 01:10:44.880 --> 01:10:48.720 align:middle line:84% happened in 2.5% of idiopathic-- that means without a known 01:10:48.720 --> 01:10:51.660 align:middle line:84% cause, short children have SHOX mutations, 01:10:51.660 --> 01:10:53.580 align:middle line:90% causing them to be short. 01:10:53.580 --> 01:10:58.130 align:middle line:84% And these mutants have perfectly normal growth hormone. 01:10:58.130 --> 01:10:59.880 align:middle line:84% But if you treat them with growth hormone, 01:10:59.880 --> 01:11:01.770 align:middle line:90% they'll grow to a normal size. 01:11:01.770 --> 01:11:06.180 align:middle line:84% And this is known in the literature and just not tested. 01:11:06.180 --> 01:11:09.120 align:middle line:84% I actually tested this because of this darn class. 01:11:09.120 --> 01:11:11.295 align:middle line:84% I made sure I was fully familiar. 01:11:11.295 --> 01:11:13.830 align:middle line:84% But like I said, in my own clinic, now probably 01:11:13.830 --> 01:11:16.110 align:middle line:84% less than 25% of the doctors order this. 01:11:16.110 --> 01:11:19.140 align:middle line:84% And yet this is a fully treatable cause 01:11:19.140 --> 01:11:20.790 align:middle line:90% of short stature. 01:11:20.790 --> 01:11:24.810 align:middle line:84% And unlike the-- and it's not as expensive 01:11:24.810 --> 01:11:28.470 align:middle line:84% as the big insulin glucagon test I gave you. 01:11:28.470 --> 01:11:30.665 align:middle line:84% And it's much more certain outcome. 01:11:30.665 --> 01:11:32.040 align:middle line:84% If you have this mutation, you're 01:11:32.040 --> 01:11:33.210 align:middle line:84% definitely going to be short, and you're definitely 01:11:33.210 --> 01:11:34.570 align:middle line:90% going to respond. 01:11:34.570 --> 01:11:36.270 align:middle line:84% So this is just a bit of a teaser 01:11:36.270 --> 01:11:39.360 align:middle line:84% to show you how today well within my reach, 01:11:39.360 --> 01:11:41.940 align:middle line:84% I can order the test today, and I do so. 01:11:41.940 --> 01:11:42.930 align:middle line:90% It's just not done. 01:11:42.930 --> 01:11:46.390 align:middle line:84% So this brings us to the other problem, which is education. 01:11:46.390 --> 01:11:48.990 align:middle line:84% If there can be thousands of mutations 01:11:48.990 --> 01:11:51.330 align:middle line:84% that are clinically significant, how in the heck 01:11:51.330 --> 01:11:54.150 align:middle line:84% are we going to teach them to our medical students who 01:11:54.150 --> 01:11:55.470 align:middle line:90% are already overloaded? 01:11:55.470 --> 01:11:58.290 align:middle line:84% And the answer, I think, and unless you 01:11:58.290 --> 01:12:00.990 align:middle line:84% can come up with a better solution, it's one or two-fold. 01:12:00.990 --> 01:12:03.810 align:middle line:84% Either create a whole bunch of experts whose only job is just 01:12:03.810 --> 01:12:06.510 align:middle line:84% to look at the patient as a doc and say, 01:12:06.510 --> 01:12:08.470 align:middle line:84% do they have X, Y, or Z, or we have 01:12:08.470 --> 01:12:11.560 align:middle line:84% to bring information technology to bear right in the process 01:12:11.560 --> 01:12:13.980 align:middle line:84% so that as you enter characteristics of patients, 01:12:13.980 --> 01:12:16.260 align:middle line:84% all the genetic possibilities start getting updated. 01:12:16.260 --> 01:12:19.250 align:middle line:90% 01:12:19.250 --> 01:12:21.950 align:middle line:84% And that's a problem because this is a decision 01:12:21.950 --> 01:12:25.850 align:middle line:84% tree for the treatment of prostate cancer based 01:12:25.850 --> 01:12:30.320 align:middle line:84% on an old proteomic test, namely prostate specific antigen. 01:12:30.320 --> 01:12:34.640 align:middle line:84% And basically, what do you do to someone who has basically 01:12:34.640 --> 01:12:38.330 align:middle line:90% a big bump in their prostate? 01:12:38.330 --> 01:12:43.575 align:middle line:84% And the problem is, if you look at the literature-- 01:12:43.575 --> 01:12:44.450 align:middle line:90% I did this last year. 01:12:44.450 --> 01:12:48.470 align:middle line:84% That's 7,325 articles, 7,000 articles-- sorry 01:12:48.470 --> 01:12:50.420 align:middle line:90% about the color or lack of it. 01:12:50.420 --> 01:12:51.890 align:middle line:84% 7,000 articles are saying, what is 01:12:51.890 --> 01:12:57.350 align:middle line:84% the right ordering of decisions based on this one test? 01:12:57.350 --> 01:13:02.220 align:middle line:84% So depending on different PSA levels. 01:13:02.220 --> 01:13:07.580 align:middle line:84% Now that's just one proteomic test for one gene. 01:13:07.580 --> 01:13:10.250 align:middle line:84% What in the heck are we going to do with these decision trees 01:13:10.250 --> 01:13:11.600 align:middle line:90% when we have to measure-- 01:13:11.600 --> 01:13:14.630 align:middle line:84% or we will measure, as we can today, thousands 01:13:14.630 --> 01:13:17.930 align:middle line:84% of gene expression levels or protein levels 01:13:17.930 --> 01:13:19.550 align:middle line:84% makes up for a much more complex tree, 01:13:19.550 --> 01:13:21.592 align:middle line:84% and we'll never have enough patients, by the way, 01:13:21.592 --> 01:13:23.390 align:middle line:84% to sample all the different possibilities. 01:13:23.390 --> 01:13:26.300 align:middle line:84% But that's a fundamental challenge of genomic medicine. 01:13:26.300 --> 01:13:29.730 align:middle line:90% 01:13:29.730 --> 01:13:31.480 align:middle line:84% So this is what I was alluded to, and just 01:13:31.480 --> 01:13:33.772 align:middle line:84% in the interest of time, I'm going to short circuit it, 01:13:33.772 --> 01:13:36.890 align:middle line:84% although this slide will be on the web shortly. 01:13:36.890 --> 01:13:39.980 align:middle line:84% If you only had 10 diseases, 10 hypotheses, 01:13:39.980 --> 01:13:41.965 align:middle line:84% and you had five binary tests-- so there's not 01:13:41.965 --> 01:13:43.340 align:middle line:84% like a prostate-specific antigen. 01:13:43.340 --> 01:13:46.030 align:middle line:84% It's not different levels in a continuous fashion and just 01:13:46.030 --> 01:13:48.060 align:middle line:90% say it's normal or abnormal. 01:13:48.060 --> 01:13:50.060 align:middle line:90% So five binary tests-- 01:13:50.060 --> 01:13:54.653 align:middle line:84% the analysis requires knowing 63,000 probabilities 01:13:54.653 --> 01:13:56.570 align:middle line:84% to be able to calculate the probability of all 01:13:56.570 --> 01:14:00.710 align:middle line:84% the possible outcomes under all the different tests. 01:14:00.710 --> 01:14:04.528 align:middle line:84% And that's just very hard to imagine 01:14:04.528 --> 01:14:06.320 align:middle line:84% how we're going to do that comprehensively. 01:14:06.320 --> 01:14:08.240 align:middle line:84% And in fact, the number of diseases 01:14:08.240 --> 01:14:09.500 align:middle line:90% is much larger than this. 01:14:09.500 --> 01:14:12.570 align:middle line:84% The number of tests that we can do is much larger than that. 01:14:12.570 --> 01:14:14.930 align:middle line:84% So that's a fundamental methodological challenge 01:14:14.930 --> 01:14:16.760 align:middle line:90% for new medicine. 01:14:16.760 --> 01:14:20.528 align:middle line:84% And in the same sense, I'm going to just-- now 01:14:20.528 --> 01:14:21.695 align:middle line:90% this will be my last slides. 01:14:21.695 --> 01:14:24.710 align:middle line:90% 01:14:24.710 --> 01:14:27.060 align:middle line:84% Evidence-based medicine is a big movement in medicine 01:14:27.060 --> 01:14:28.727 align:middle line:84% which says, let's look at the literature 01:14:28.727 --> 01:14:30.590 align:middle line:84% and try to find out what the best practice. 01:14:30.590 --> 01:14:32.930 align:middle line:84% And they've done this for a number of different disease 01:14:32.930 --> 01:14:36.950 align:middle line:84% areas, and there's literally thousands 01:14:36.950 --> 01:14:39.800 align:middle line:84% of doctors and researchers who are filling out 01:14:39.800 --> 01:14:43.050 align:middle line:84% these databases of evidence-based medicine, 01:14:43.050 --> 01:14:45.380 align:middle line:84% the largest being the Cochrane collaborative. 01:14:45.380 --> 01:14:51.770 align:middle line:84% But if you look at the genetic section, it's just a few 01:14:51.770 --> 01:14:54.140 align:middle line:84% highly penetrant genetic diseases. 01:14:54.140 --> 01:14:55.910 align:middle line:84% There's very little to guide-- in fact, 01:14:55.910 --> 01:14:59.600 align:middle line:84% there's nothing to guide the practitioner for anything 01:14:59.600 --> 01:15:03.140 align:middle line:84% of these genomic diseases for these very common diseases, 01:15:03.140 --> 01:15:08.360 align:middle line:84% which is where I stop today's class 01:15:08.360 --> 01:15:10.040 align:middle line:90% because we ran out of time. 01:15:10.040 --> 01:15:12.290 align:middle line:84% And for another time, I'll tell you 01:15:12.290 --> 01:15:13.970 align:middle line:84% about the problems about patients 01:15:13.970 --> 01:15:20.270 align:middle line:84% who are able to select tests, order tests over the web today. 01:15:20.270 --> 01:15:24.260 align:middle line:84% You can actually send your own samples to be sequenced 01:15:24.260 --> 01:15:28.280 align:middle line:84% and a dubious interpretation or undubious interpretation 01:15:28.280 --> 01:15:29.120 align:middle line:90% is given to you. 01:15:29.120 --> 01:15:30.890 align:middle line:84% And then you show up in a doctor's office, 01:15:30.890 --> 01:15:32.420 align:middle line:84% and he doesn't know or she doesn't 01:15:32.420 --> 01:15:33.910 align:middle line:90% know what to do with it. 01:15:33.910 --> 01:15:35.870 align:middle line:90% But that'll be for another time. 01:15:35.870 --> 01:15:42.070 align:middle line:84% All right, any administrative issues, feel free to--