WEBVTT 00:00:00.000 --> 00:00:00.610 align:middle line:90% 00:00:00.610 --> 00:00:02.440 align:middle line:84% The following content is provided 00:00:02.440 --> 00:00:06.340 align:middle line:84% by MIT OpenCourseWare under a Creative Commons license. 00:00:06.340 --> 00:00:08.470 align:middle line:84% Additional information about our license, 00:00:08.470 --> 00:00:10.870 align:middle line:84% and MIT OpenCourseWare in general, 00:00:10.870 --> 00:00:13.090 align:middle line:90% is available at ocw.mit.edu. 00:00:13.090 --> 00:00:16.610 align:middle line:90% 00:00:16.610 --> 00:00:19.830 align:middle line:84% GEORGE CHURCH: We've come to the fifth lecture, 00:00:19.830 --> 00:00:26.660 align:middle line:84% the first of a series, on RNA and expression analysis. 00:00:26.660 --> 00:00:30.320 align:middle line:84% First, a quick review of last week, and it 00:00:30.320 --> 00:00:32.374 align:middle line:84% has significant connections this week, 00:00:32.374 --> 00:00:37.090 align:middle line:84% where we talked to the topic of alignments 00:00:37.090 --> 00:00:40.750 align:middle line:84% and different algorithms for obtaining pairwise alignments. 00:00:40.750 --> 00:00:43.060 align:middle line:84% In particular dynamic programming, 00:00:43.060 --> 00:00:45.470 align:middle line:84% this led to an even harder problem, 00:00:45.470 --> 00:00:47.770 align:middle line:84% which is multi-sequence alignment, from which we 00:00:47.770 --> 00:00:52.620 align:middle line:84% will draw pretty heavily at the beginning of the discussion 00:00:52.620 --> 00:00:53.760 align:middle line:90% today. 00:00:53.760 --> 00:00:59.780 align:middle line:84% And then, the issues of getting the motifs, 00:00:59.780 --> 00:01:02.390 align:middle line:84% another topic we will touch upon several times today, 00:01:02.390 --> 00:01:05.120 align:middle line:84% how you get-- once you have a multi-sequence alignment, how 00:01:05.120 --> 00:01:10.430 align:middle line:84% this gives you either an independent weight matrix where 00:01:10.430 --> 00:01:13.610 align:middle line:84% different positions are independent or in a hidden 00:01:13.610 --> 00:01:17.570 align:middle line:84% Markov model where there is some dependence between, say, 00:01:17.570 --> 00:01:21.515 align:middle line:84% adjacent nucleotides in a simple sequence such as CG. 00:01:21.515 --> 00:01:24.150 align:middle line:90% 00:01:24.150 --> 00:01:27.780 align:middle line:84% So let us carry over these thoughts 00:01:27.780 --> 00:01:32.370 align:middle line:84% about multi-sequence alignment motifs 00:01:32.370 --> 00:01:36.960 align:middle line:84% and non independence of positions in sequences 00:01:36.960 --> 00:01:39.540 align:middle line:90% to the next level higher. 00:01:39.540 --> 00:01:41.193 align:middle line:84% We will eventually talk about protein 00:01:41.193 --> 00:01:42.360 align:middle line:90% three-dimensional structure. 00:01:42.360 --> 00:01:48.160 align:middle line:84% But a really beautiful intermediate between proteins 00:01:48.160 --> 00:01:50.290 align:middle line:84% are the complexities of protein structure 00:01:50.290 --> 00:01:53.350 align:middle line:84% and the simplicity of double-stranded DNA 00:01:53.350 --> 00:01:59.590 align:middle line:84% is the folding of RNAs because they're 00:01:59.590 --> 00:02:01.660 align:middle line:84% based on the same rules of double-stranded DNA, 00:02:01.660 --> 00:02:06.320 align:middle line:84% but they have the complicated structures 00:02:06.320 --> 00:02:09.000 align:middle line:84% of-- begin to have a complicated structure of proteins. 00:02:09.000 --> 00:02:11.870 align:middle line:84% So we'll start with this integration 00:02:11.870 --> 00:02:16.700 align:middle line:84% of multi-sequence alignment of motifs with RNA structure, 00:02:16.700 --> 00:02:23.180 align:middle line:84% and then we'll switch to tell about how these RNA 00:02:23.180 --> 00:02:27.170 align:middle line:84% structures play their role by achieving 00:02:27.170 --> 00:02:30.110 align:middle line:90% different levels in the cell. 00:02:30.110 --> 00:02:32.750 align:middle line:84% In other words, we want to start to introduce 00:02:32.750 --> 00:02:37.220 align:middle line:84% how we become quantitative about the amounts and localization 00:02:37.220 --> 00:02:38.900 align:middle line:90% of RNAs in the cell. 00:02:38.900 --> 00:02:42.500 align:middle line:84% Some of the measures that we'll be talking about and the tool-- 00:02:42.500 --> 00:02:47.960 align:middle line:84% the computational tools will be more appropriate 00:02:47.960 --> 00:02:50.660 align:middle line:84% for individual measures and others will be more, 00:02:50.660 --> 00:02:54.380 align:middle line:84% what we call, genomics grade, high throughput 00:02:54.380 --> 00:02:57.780 align:middle line:90% and high accuracy. 00:02:57.780 --> 00:03:00.770 align:middle line:90% And we have this-- 00:03:00.770 --> 00:03:06.590 align:middle line:84% since this is a new category of biological data, 00:03:06.590 --> 00:03:09.320 align:middle line:84% we have to address random and systematic errors 00:03:09.320 --> 00:03:12.290 align:middle line:84% just as we did for genotyping and sequence data. 00:03:12.290 --> 00:03:14.720 align:middle line:84% This is a new set of them and a new set 00:03:14.720 --> 00:03:18.120 align:middle line:84% of solutions about the same themes of random and systematic 00:03:18.120 --> 00:03:18.620 align:middle line:90% errors. 00:03:18.620 --> 00:03:22.250 align:middle line:84% And then we'll talk about a particular set 00:03:22.250 --> 00:03:23.930 align:middle line:84% of interpretation issues that lead 00:03:23.930 --> 00:03:26.510 align:middle line:90% to additional considerations. 00:03:26.510 --> 00:03:29.420 align:middle line:84% And we'll end on time series data which 00:03:29.420 --> 00:03:31.700 align:middle line:84% will be a theme that will connect 00:03:31.700 --> 00:03:34.940 align:middle line:84% this talk to much later talks and systems biology 00:03:34.940 --> 00:03:39.110 align:middle line:84% where the ability to connect with a time series 00:03:39.110 --> 00:03:46.550 align:middle line:84% will help establish causality and connectivity. 00:03:46.550 --> 00:03:49.530 align:middle line:84% And we'll tie it to the subject of RNA analysis 00:03:49.530 --> 00:03:52.480 align:middle line:84% by looking at messenger RNA to k. 00:03:52.480 --> 00:03:56.320 align:middle line:84% Now slide three is a reminder that we'll 00:03:56.320 --> 00:03:58.150 align:middle line:84% use in two different contexts tonight. 00:03:58.150 --> 00:04:01.900 align:middle line:84% First, these are the bell curves that you've seen, 00:04:01.900 --> 00:04:04.660 align:middle line:84% at least three of them integrated before-- 00:04:04.660 --> 00:04:08.200 align:middle line:84% the two discrete binomial Poisson 00:04:08.200 --> 00:04:12.250 align:middle line:84% and the normal, which is symmetric around the mean 00:04:12.250 --> 00:04:13.480 align:middle line:90% of 20, in this case. 00:04:13.480 --> 00:04:16.290 align:middle line:90% 00:04:16.290 --> 00:04:23.370 align:middle line:84% And just to connect to the discussion from last time where 00:04:23.370 --> 00:04:29.810 align:middle line:84% we were asking what the significance of a match 00:04:29.810 --> 00:04:32.720 align:middle line:84% of a single sequence to a database might be, 00:04:32.720 --> 00:04:35.660 align:middle line:84% when you're asking for a match of a single-sequence database, 00:04:35.660 --> 00:04:38.870 align:middle line:84% you're typically really asking for the maximum match, 00:04:38.870 --> 00:04:40.340 align:middle line:90% or the maximum-- 00:04:40.340 --> 00:04:43.700 align:middle line:90% the most extreme matches. 00:04:43.700 --> 00:04:45.590 align:middle line:84% And so when you talk about extremes, 00:04:45.590 --> 00:04:47.810 align:middle line:84% when you're sampling from a distribution, 00:04:47.810 --> 00:04:50.180 align:middle line:84% and you're looking for the most extreme value 00:04:50.180 --> 00:04:53.090 align:middle line:84% for finite sampling of that distribution, 00:04:53.090 --> 00:04:55.910 align:middle line:84% that tends to be not from the normal distribution, which 00:04:55.910 --> 00:04:57.920 align:middle line:84% would be random sampling, but from-- which 00:04:57.920 --> 00:05:06.470 align:middle line:90% is this middle magenta curve. 00:05:06.470 --> 00:05:09.115 align:middle line:84% But instead, it would be extreme value distribution, which 00:05:09.115 --> 00:05:11.240 align:middle line:84% is this blue curve, which you can see in this case, 00:05:11.240 --> 00:05:13.180 align:middle line:84% since we're looking for extreme maximum, 00:05:13.180 --> 00:05:15.170 align:middle line:84% it's shifted slightly to the right. 00:05:15.170 --> 00:05:19.840 align:middle line:84% And so you can see that it comes inside of the other bell 00:05:19.840 --> 00:05:22.270 align:middle line:84% curves, on the left-hand side, and goes 00:05:22.270 --> 00:05:24.010 align:middle line:90% outside, on the right-hand side. 00:05:24.010 --> 00:05:27.910 align:middle line:84% If we were looking for extremely low values then 00:05:27.910 --> 00:05:29.810 align:middle line:84% it would've been shifted to the left. 00:05:29.810 --> 00:05:34.030 align:middle line:84% And remember, all these continuous functions go off 00:05:34.030 --> 00:05:35.950 align:middle line:84% to negative infinity and positive infinity, 00:05:35.950 --> 00:05:37.390 align:middle line:84% although at extremely low levels. 00:05:37.390 --> 00:05:41.810 align:middle line:90% 00:05:41.810 --> 00:05:46.100 align:middle line:84% And then that's the extreme value distribution. 00:05:46.100 --> 00:05:52.280 align:middle line:84% Now in order to connect this nucleotide sequence, which 00:05:52.280 --> 00:05:55.310 align:middle line:84% we've been seeing has these wonderful Watson Crick base 00:05:55.310 --> 00:05:58.380 align:middle line:84% pairs that were TNA, and so forth, 00:05:58.380 --> 00:06:02.810 align:middle line:84% is to this more complicated tertiary structure. 00:06:02.810 --> 00:06:04.790 align:middle line:84% We're going to go through an intermediate 00:06:04.790 --> 00:06:06.710 align:middle line:84% of secondary structure where we really 00:06:06.710 --> 00:06:10.910 align:middle line:84% look at whether what kind of base pairs can form. 00:06:10.910 --> 00:06:14.340 align:middle line:84% And I'm going to immediately introduce some complexities 00:06:14.340 --> 00:06:20.270 align:middle line:84% so you don't get too complacent right off the bat. 00:06:20.270 --> 00:06:24.230 align:middle line:84% Somewhere in this slide of non-Watson-Crick base pairs 00:06:24.230 --> 00:06:27.940 align:middle line:90% is a Watson-Crick base pair. 00:06:27.940 --> 00:06:30.040 align:middle line:90% Take a moment to find it. 00:06:30.040 --> 00:06:30.570 align:middle line:90% Fine. 00:06:30.570 --> 00:06:31.070 align:middle line:90% OK. 00:06:31.070 --> 00:06:33.490 align:middle line:84% So since we haven't introduced base pairs, it's going to-- 00:06:33.490 --> 00:06:36.970 align:middle line:84% well, we've seen it twice now in double-stranded DNA. 00:06:36.970 --> 00:06:38.635 align:middle line:84% Right in the middle here are labeled 00:06:38.635 --> 00:06:43.150 align:middle line:84% A is an AU base pair, which is Watson-Crick 00:06:43.150 --> 00:06:47.980 align:middle line:84% where the black dot indicates the attachment 00:06:47.980 --> 00:06:52.390 align:middle line:84% to the ribose in RNA, or to the oxyribose in DNA. 00:06:52.390 --> 00:06:55.210 align:middle line:84% And you'll find three other AT base pairs. 00:06:55.210 --> 00:06:59.560 align:middle line:84% One right to the right of it and two down below. 00:06:59.560 --> 00:07:02.590 align:middle line:84% And these four AT base pairs are all 00:07:02.590 --> 00:07:07.940 align:middle line:84% different from one another, most easily 00:07:07.940 --> 00:07:09.710 align:middle line:84% imagined in terms of the orientation 00:07:09.710 --> 00:07:14.670 align:middle line:84% of the riboses, these black dots, relative to one another. 00:07:14.670 --> 00:07:16.370 align:middle line:90% And they each have names. 00:07:16.370 --> 00:07:18.790 align:middle line:84% But the important thing is that all of these 00:07:18.790 --> 00:07:22.190 align:middle line:84% are illustrated such that they maintain 00:07:22.190 --> 00:07:24.380 align:middle line:90% the coplanarity of the bases. 00:07:24.380 --> 00:07:27.080 align:middle line:84% They typically maintain one, or two, or even 00:07:27.080 --> 00:07:29.510 align:middle line:90% three hydrogen bonds. 00:07:29.510 --> 00:07:34.130 align:middle line:84% And the planarity allows them to stack on base pairs 00:07:34.130 --> 00:07:36.650 align:middle line:84% below and above them, just as you would 00:07:36.650 --> 00:07:38.255 align:middle line:90% in normal double-stranded DNA. 00:07:38.255 --> 00:07:41.040 align:middle line:90% 00:07:41.040 --> 00:07:43.680 align:middle line:84% But sometimes, the geometry distorts the double helix 00:07:43.680 --> 00:07:46.020 align:middle line:84% enough that you might get a penalty 00:07:46.020 --> 00:07:50.910 align:middle line:84% in the free energy in the thermodynamic sense, 00:07:50.910 --> 00:07:54.030 align:middle line:90% or the kinetics, in a sense. 00:07:54.030 --> 00:07:55.000 align:middle line:90% Now here's another. 00:07:55.000 --> 00:07:57.180 align:middle line:84% So you can find almost-- you can basically 00:07:57.180 --> 00:07:59.010 align:middle line:84% make one of these base pairs for all 00:07:59.010 --> 00:08:09.280 align:middle line:84% of the possible combinations of ACG and U in RNA. 00:08:09.280 --> 00:08:10.660 align:middle line:84% And all of them will be coplanar, 00:08:10.660 --> 00:08:13.990 align:middle line:84% and they will have one or more hydrogen bonds. 00:08:13.990 --> 00:08:16.330 align:middle line:84% Probably the most stable and most commonly 00:08:16.330 --> 00:08:19.810 align:middle line:84% encountered in otherwise normal RNA 00:08:19.810 --> 00:08:22.370 align:middle line:84% double helices is a GU base pair. 00:08:22.370 --> 00:08:25.210 align:middle line:84% And you can see that this has fairly similar geometry 00:08:25.210 --> 00:08:30.500 align:middle line:84% to the AU base pair, or for that matter, the GC base pair. 00:08:30.500 --> 00:08:35.370 align:middle line:84% So let's see how these non Watson-Crick base pairs appear. 00:08:35.370 --> 00:08:39.470 align:middle line:84% And this is the transfer RNA that we 00:08:39.470 --> 00:08:41.809 align:middle line:84% saw a couple of slides ago, spinning around 00:08:41.809 --> 00:08:42.740 align:middle line:90% in three dimensions. 00:08:42.740 --> 00:08:46.190 align:middle line:84% And the sequence that was behind it 00:08:46.190 --> 00:08:49.700 align:middle line:84% was the DNA sequence that corresponds 00:08:49.700 --> 00:08:53.015 align:middle line:84% to this unmodified RNA sequence on the right-hand side. 00:08:53.015 --> 00:08:56.340 align:middle line:90% 00:08:56.340 --> 00:09:02.550 align:middle line:84% What we have-- you can think of this as four fairly canonical 00:09:02.550 --> 00:09:07.650 align:middle line:84% Watson-Crick type double helices for RNA double helices, which 00:09:07.650 --> 00:09:12.380 align:middle line:84% are slightly different from the DNA double helices. 00:09:12.380 --> 00:09:16.940 align:middle line:84% But in this-- so we have seven base pairs, six of which 00:09:16.940 --> 00:09:20.690 align:middle line:84% are Watson-Crick in the top stem loop, 00:09:20.690 --> 00:09:23.690 align:middle line:84% starting at position number one with the 5-prime end 00:09:23.690 --> 00:09:26.370 align:middle line:90% and ending position 72. 00:09:26.370 --> 00:09:29.250 align:middle line:90% So one and 72 is a GC base pair. 00:09:29.250 --> 00:09:31.760 align:middle line:84% And you can see that the anticodon where 00:09:31.760 --> 00:09:35.450 align:middle line:84% it meets the messenger RNA is at the bottom right-hand part 00:09:35.450 --> 00:09:36.950 align:middle line:90% of the slide. 00:09:36.950 --> 00:09:39.920 align:middle line:84% And if you just look at that loop and-- 00:09:39.920 --> 00:09:45.260 align:middle line:84% you've got a seven-base loop and a five-base pair anticodon 00:09:45.260 --> 00:09:46.250 align:middle line:90% stem. 00:09:46.250 --> 00:09:47.310 align:middle line:90% And so each of these-- 00:09:47.310 --> 00:09:51.830 align:middle line:84% so that each of these stems has some distinguishing features, 00:09:51.830 --> 00:09:55.370 align:middle line:84% a number of base pairs ranges from four to seven. 00:09:55.370 --> 00:09:59.360 align:middle line:84% You've got a GU base pair in the middle of the top stem. 00:09:59.360 --> 00:10:02.000 align:middle line:84% And you've got little sequence boxes 00:10:02.000 --> 00:10:09.340 align:middle line:84% which are fairly conserved, such as this T psi CG The-- 00:10:09.340 --> 00:10:13.510 align:middle line:84% is-- actually, in its original form is a UU CG. 00:10:13.510 --> 00:10:17.470 align:middle line:84% And the T and the psi, or pseudouridine, 00:10:17.470 --> 00:10:19.480 align:middle line:84% are examples of modified bases, which are shown 00:10:19.480 --> 00:10:21.110 align:middle line:90% on the left side of the slide. 00:10:21.110 --> 00:10:23.320 align:middle line:84% You can see there's quite a number of them. 00:10:23.320 --> 00:10:27.340 align:middle line:84% You can add a methyl group of ch3 groups 00:10:27.340 --> 00:10:30.940 align:middle line:84% to either of the bases, such as one methyl group on G. 00:10:30.940 --> 00:10:33.640 align:middle line:84% Or you can add them to the riboses, such as the two 00:10:33.640 --> 00:10:38.170 align:middle line:84% prime methyl groups, which can go on any of the four bases 00:10:38.170 --> 00:10:41.350 align:middle line:84% because it's modifying the ribose which is generic. 00:10:41.350 --> 00:10:44.110 align:middle line:84% Most of the other ones are very specific to a base. 00:10:44.110 --> 00:10:46.630 align:middle line:84% So for example, dihydrouridine is a modification that 00:10:46.630 --> 00:10:49.210 align:middle line:90% can only occur to uridines. 00:10:49.210 --> 00:10:52.285 align:middle line:84% The pseudouridine-- similar thing. 00:10:52.285 --> 00:10:55.660 align:middle line:90% 00:10:55.660 --> 00:10:57.180 align:middle line:90% And so on. 00:10:57.180 --> 00:10:59.640 align:middle line:84% So each of these requires an enzyme. 00:10:59.640 --> 00:11:02.970 align:middle line:84% And we will highlight one of the enzymes that's involved 00:11:02.970 --> 00:11:07.140 align:middle line:84% in putting the methyl groups onto the sugars, 00:11:07.140 --> 00:11:10.210 align:middle line:84% the [? 0,2 ?] [? trimethyl ?] in just a few slides. 00:11:10.210 --> 00:11:11.880 align:middle line:84% But right now, what we want to ask 00:11:11.880 --> 00:11:13.930 align:middle line:84% is how did we get this folding structure. 00:11:13.930 --> 00:11:15.930 align:middle line:84% Now this is not the three-dimensional structure. 00:11:15.930 --> 00:11:18.570 align:middle line:84% This is the intermediate, between the primary DNA 00:11:18.570 --> 00:11:21.738 align:middle line:84% sequence and the fully modified, fully folded 00:11:21.738 --> 00:11:24.030 align:middle line:84% three-dimensional structure that we saw spinning around 00:11:24.030 --> 00:11:25.450 align:middle line:90% a couple slides ago. 00:11:25.450 --> 00:11:31.250 align:middle line:84% So the first thing is the way-- you can try folding 00:11:31.250 --> 00:11:36.170 align:middle line:84% this, for you oppose each base pair in turn 00:11:36.170 --> 00:11:38.540 align:middle line:90% to look for possible matches. 00:11:38.540 --> 00:11:41.150 align:middle line:84% And then, what we've done historically-- 00:11:41.150 --> 00:11:43.640 align:middle line:84% this is in the mid to late '60s-- 00:11:43.640 --> 00:11:46.190 align:middle line:84% was you take each new transfer RNA sequence 00:11:46.190 --> 00:11:49.880 align:middle line:84% and ask whether it makes a decent fold 00:11:49.880 --> 00:11:52.280 align:middle line:84% in this simple planar representation 00:11:52.280 --> 00:11:54.320 align:middle line:84% that's related to the previous ones, 00:11:54.320 --> 00:11:57.260 align:middle line:84% under the assumption, the hope that there would 00:11:57.260 --> 00:12:01.040 align:middle line:84% be some conservation, not only of sequence of some 00:12:01.040 --> 00:12:05.600 align:middle line:84% of these motifs like the PSI CG, but also the way that it 00:12:05.600 --> 00:12:06.290 align:middle line:90% folds up. 00:12:06.290 --> 00:12:08.600 align:middle line:84% And you might even hypothesize that maybe it 00:12:08.600 --> 00:12:11.180 align:middle line:84% doesn't matter what the sequence is in some of these stems. 00:12:11.180 --> 00:12:13.940 align:middle line:84% What's important is that it's capable of forming a stem, 00:12:13.940 --> 00:12:19.370 align:middle line:84% that it is that position one is complementary to position 72. 00:12:19.370 --> 00:12:22.010 align:middle line:84% If position one were to change from a G to an A, 00:12:22.010 --> 00:12:26.840 align:middle line:84% then position 72 should change from a C to a U. 00:12:26.840 --> 00:12:31.050 align:middle line:90% So how do we formalize this? 00:12:31.050 --> 00:12:34.740 align:middle line:84% How do we formalize the process by which we generate 00:12:34.740 --> 00:12:36.420 align:middle line:84% this so-called cloverleaf structural 00:12:36.420 --> 00:12:43.670 align:middle line:84% or any similar folding pattern for small nucleic acids. 00:12:43.670 --> 00:12:46.880 align:middle line:84% And what are the limitations of those algorithms. 00:12:46.880 --> 00:12:49.310 align:middle line:84% And these dotted lines, you see, are 00:12:49.310 --> 00:12:51.470 align:middle line:84% some of the non Watson-Crick based pairs. 00:12:51.470 --> 00:12:52.820 align:middle line:90% Some of them will stack. 00:12:52.820 --> 00:12:54.590 align:middle line:84% Many of them-- some of them will actually 00:12:54.590 --> 00:12:58.280 align:middle line:84% form the hydrogen bonds of the Watson-Crick based pair 00:12:58.280 --> 00:13:02.210 align:middle line:84% but they won't otherwise have the rest of the geometry. 00:13:02.210 --> 00:13:03.830 align:middle line:84% And you can see that some of these 00:13:03.830 --> 00:13:06.740 align:middle line:84% will provide connections between two loops which 00:13:06.740 --> 00:13:08.180 align:middle line:90% are separated by stems. 00:13:08.180 --> 00:13:13.580 align:middle line:84% And this kind of folding back means that it's not 00:13:13.580 --> 00:13:18.170 align:middle line:90% a simple set of helices. 00:13:18.170 --> 00:13:20.240 align:middle line:84% So the way that we formalize this 00:13:20.240 --> 00:13:24.980 align:middle line:84% is we say that position number one, if it's 00:13:24.980 --> 00:13:31.060 align:middle line:84% bound to position number 72, and the exact sequence maybe 00:13:31.060 --> 00:13:34.030 align:middle line:84% isn't as important as their ability to pair well another, 00:13:34.030 --> 00:13:37.150 align:middle line:84% then you expect if you take a large number of transfer RNAs 00:13:37.150 --> 00:13:40.420 align:middle line:84% and do a multi-sequence alignment, as we did last time, 00:13:40.420 --> 00:13:44.710 align:middle line:84% then in that multi-sequence alignment, 00:13:44.710 --> 00:13:48.610 align:middle line:84% you expect that when the G changes, 00:13:48.610 --> 00:13:49.600 align:middle line:90% the C will change, too. 00:13:49.600 --> 00:13:51.100 align:middle line:90% And that's called covariance. 00:13:51.100 --> 00:13:53.290 align:middle line:84% And if you look at the vertical axis, 00:13:53.290 --> 00:13:58.200 align:middle line:84% the maximum that can be achieved is the same kind of maximum 00:13:58.200 --> 00:14:01.440 align:middle line:84% that we had in the motifs last lecture. 00:14:01.440 --> 00:14:04.800 align:middle line:84% The motifs, we've actually had a couple of times. 00:14:04.800 --> 00:14:06.270 align:middle line:90% It can get up to two bits. 00:14:06.270 --> 00:14:11.310 align:middle line:84% Two bits is the full scale for a base pair, 00:14:11.310 --> 00:14:14.130 align:middle line:84% or base, which can have four different values, A, C, 00:14:14.130 --> 00:14:16.740 align:middle line:84% G, or T. And we're calling this mutual information 00:14:16.740 --> 00:14:20.830 align:middle line:84% if it has the same units, a full scale of zero to two bits. 00:14:20.830 --> 00:14:25.140 align:middle line:84% And so we see along the horizontal axis, here-- 00:14:25.140 --> 00:14:29.970 align:middle line:84% call them positions I and J, which range from one to 72, 00:14:29.970 --> 00:14:32.040 align:middle line:84% which is the core part of the transfer RNA. 00:14:32.040 --> 00:14:37.560 align:middle line:84% The last four bases are added by a specialized enzyme. 00:14:37.560 --> 00:14:40.200 align:middle line:84% But position number one and position 72 00:14:40.200 --> 00:14:41.910 align:middle line:84% covariates [? assume ?] this peak 00:14:41.910 --> 00:14:46.390 align:middle line:90% in the far left-hand region. 00:14:46.390 --> 00:14:51.060 align:middle line:84% In effect, there's seven peaks in a row there 00:14:51.060 --> 00:14:56.430 align:middle line:84% which correspond to the seven stacked base pairs they covary. 00:14:56.430 --> 00:15:01.320 align:middle line:84% Similarly, in the TC to UC stem that we talked about a couple 00:15:01.320 --> 00:15:06.270 align:middle line:84% of times there, that those five nucleotides covary 00:15:06.270 --> 00:15:12.390 align:middle line:90% as you would get in a stem. 00:15:12.390 --> 00:15:15.220 align:middle line:84% The anti-codon stem is another five. 00:15:15.220 --> 00:15:20.160 align:middle line:84% And that D stem, so named after the hydrourodine modifications 00:15:20.160 --> 00:15:21.690 align:middle line:90% is four base pairs. 00:15:21.690 --> 00:15:23.280 align:middle line:84% And the way that this is derived-- 00:15:23.280 --> 00:15:25.510 align:middle line:84% and we're going to work through this, an example, 00:15:25.510 --> 00:15:31.660 align:middle line:84% in the next slide-- but just as a labeling of this axis here, 00:15:31.660 --> 00:15:36.130 align:middle line:84% the mutual information between the Ith base and the jth base, 00:15:36.130 --> 00:15:39.430 align:middle line:84% that is to see, for example, between I equals 1 and J equals 00:15:39.430 --> 00:15:44.800 align:middle line:84% 72, is simply the sum of the frequency of getting that 00:15:44.800 --> 00:15:46.960 align:middle line:90% particular I, N, J-- 00:15:46.960 --> 00:15:51.320 align:middle line:84% F is the frequency of getting, say, 00:15:51.320 --> 00:15:56.500 align:middle line:84% a G as position one and a C at position 72-- 00:15:56.500 --> 00:15:58.453 align:middle line:90% times the log base 2. 00:15:58.453 --> 00:16:00.370 align:middle line:84% Remember, when we're talking about information 00:16:00.370 --> 00:16:03.070 align:middle line:90% content of nuclear-- 00:16:03.070 --> 00:16:06.310 align:middle line:84% or of information in general, bits in a computer 00:16:06.310 --> 00:16:11.080 align:middle line:84% or nucleotides in a sequence, we do log base 2, and-- 00:16:11.080 --> 00:16:12.830 align:middle line:84% as introduced by Shannon and others. 00:16:12.830 --> 00:16:15.830 align:middle line:84% So now it's going to be the log of that same frequency. 00:16:15.830 --> 00:16:19.420 align:middle line:84% So the frequency of getting that particular I 00:16:19.420 --> 00:16:23.170 align:middle line:84% and that particular J type of base 00:16:23.170 --> 00:16:28.450 align:middle line:84% normalized now to how frequently those two bases occur 00:16:28.450 --> 00:16:33.800 align:middle line:90% throughout those positions. 00:16:33.800 --> 00:16:36.700 align:middle line:84% In other words, you know how often they co-occur 00:16:36.700 --> 00:16:38.660 align:middle line:90% at those two positions. 00:16:38.660 --> 00:16:41.300 align:middle line:84% Now, how often do they occur independently of one another? 00:16:41.300 --> 00:16:43.790 align:middle line:84% And that's what the denominator is here. 00:16:43.790 --> 00:16:49.420 align:middle line:84% So when you take this ratio, you put it on a conventional scale, 00:16:49.420 --> 00:16:51.070 align:middle line:84% and then you have something that's 00:16:51.070 --> 00:16:53.560 align:middle line:84% analogous to the P law P of information theory. 00:16:53.560 --> 00:17:01.220 align:middle line:84% And you sum over all of the observed bases 00:17:01.220 --> 00:17:06.220 align:middle line:90% at positions I and J. 00:17:06.220 --> 00:17:08.859 align:middle line:84% And that that's for a particular I and J, 00:17:08.859 --> 00:17:12.940 align:middle line:84% you sum over all the X's that occur at position, say, one 00:17:12.940 --> 00:17:14.170 align:middle line:90% and 72. 00:17:14.170 --> 00:17:16.069 align:middle line:90% And then you repeat that. 00:17:16.069 --> 00:17:21.530 align:middle line:84% You can get this Nth of IJ for every matrix element 00:17:21.530 --> 00:17:28.700 align:middle line:84% going from one to 72 in a symmetric square matrix. 00:17:28.700 --> 00:17:31.640 align:middle line:84% So let's work through this for two extreme examples. 00:17:31.640 --> 00:17:34.940 align:middle line:84% The extreme case where have perfect covariance 00:17:34.940 --> 00:17:39.350 align:middle line:84% and the extreme case where you have no real association. 00:17:39.350 --> 00:17:42.800 align:middle line:84% So we're going to illustrate this with a toy multi-sequence 00:17:42.800 --> 00:17:43.670 align:middle line:90% alignment here. 00:17:43.670 --> 00:17:47.120 align:middle line:84% This is just the same way we did the multi-sequence alignment 00:17:47.120 --> 00:17:48.020 align:middle line:90% in the last class. 00:17:48.020 --> 00:17:50.180 align:middle line:84% Here there are no insertions or deletions, 00:17:50.180 --> 00:17:51.590 align:middle line:90% but it's the same thing. 00:17:51.590 --> 00:17:53.990 align:middle line:84% You could derive a weight matrix for this. 00:17:53.990 --> 00:17:56.570 align:middle line:84% And you would see that the first column-- the far left-hand 00:17:56.570 --> 00:18:00.740 align:middle line:84% column, I equals one, has all four possibilities, 00:18:00.740 --> 00:18:04.970 align:middle line:84% and the so does the rightmost column 00:18:04.970 --> 00:18:08.300 align:middle line:84% of that multi-sequence alignment J equals six. 00:18:08.300 --> 00:18:13.310 align:middle line:84% And so let's calculate, are these covarying 00:18:13.310 --> 00:18:19.210 align:middle line:84% in this simple multi-sequence alignment of four [INAUDIBLE].. 00:18:19.210 --> 00:18:21.940 align:middle line:84% So we calculate the mutual information for I equals one, 00:18:21.940 --> 00:18:25.420 align:middle line:90% J equals six Nth of one sixth. 00:18:25.420 --> 00:18:28.210 align:middle line:90% It's going to be equal to sum. 00:18:28.210 --> 00:18:31.660 align:middle line:84% The first term in the sum is for the AU. 00:18:31.660 --> 00:18:34.600 align:middle line:84% And then we're going to walk through the CG, GC, UA. 00:18:34.600 --> 00:18:36.940 align:middle line:84% So there will be four terms in the sum. 00:18:36.940 --> 00:18:40.270 align:middle line:84% Each of the terms will have, coincidentally in this case, 00:18:40.270 --> 00:18:43.720 align:middle line:84% have the same frequency for that particular pairing of AU. 00:18:43.720 --> 00:18:45.350 align:middle line:84% And remember this is not a base pair. 00:18:45.350 --> 00:18:47.290 align:middle line:84% This is a covariant pair of nucleotides 00:18:47.290 --> 00:18:49.207 align:middle line:84% that could have been anywhere in the sequence. 00:18:49.207 --> 00:18:51.475 align:middle line:84% We happen to pick the first and the last base. 00:18:51.475 --> 00:18:54.040 align:middle line:90% 00:18:54.040 --> 00:18:55.780 align:middle line:84% So they all have the same frequency. 00:18:55.780 --> 00:18:57.070 align:middle line:90% That frequency is 1/4. 00:18:57.070 --> 00:18:59.470 align:middle line:84% So the AU occurs one quarter of the 00:18:59.470 --> 00:19:01.580 align:middle line:84% for sequences in the multi-sequence alignment, 00:19:01.580 --> 00:19:02.890 align:middle line:90% so that's one quarter. 00:19:02.890 --> 00:19:07.570 align:middle line:84% And then, remember it's the same frequency inside the logarithm, 00:19:07.570 --> 00:19:09.640 align:middle line:84% but now in the denominator, we're 00:19:09.640 --> 00:19:12.100 align:middle line:84% going to normalize it to the frequency 00:19:12.100 --> 00:19:16.090 align:middle line:84% that the A occurs in the I equals one position, which 00:19:16.090 --> 00:19:17.800 align:middle line:84% is one quarter, and the frequency that U 00:19:17.800 --> 00:19:20.770 align:middle line:84% occurs in the J equals six position, which is one quarter. 00:19:20.770 --> 00:19:25.090 align:middle line:84% So that's one quarter over one quarter squared, or four. 00:19:25.090 --> 00:19:30.400 align:middle line:84% So 0.25 times log base to a of four is going to be two. 00:19:30.400 --> 00:19:39.910 align:middle line:84% And 0.25 times two is going to be 0.5. 00:19:39.910 --> 00:19:43.150 align:middle line:90% And that's the first term. 00:19:43.150 --> 00:19:45.130 align:middle line:90% That's for the AU pairing. 00:19:45.130 --> 00:19:46.870 align:middle line:84% If you go down through all four terms, 00:19:46.870 --> 00:19:49.990 align:middle line:84% they all end up being the same form. 00:19:49.990 --> 00:19:54.070 align:middle line:84% The frequency is always going to be 0.25 for the pair and 0.25 00:19:54.070 --> 00:19:55.720 align:middle line:84% for each of the individual bases. 00:19:55.720 --> 00:19:59.290 align:middle line:84% So you end up with four of those, four examples of those 00:19:59.290 --> 00:20:01.360 align:middle line:90% for each of the four cases. 00:20:01.360 --> 00:20:04.930 align:middle line:90% And so four times 0.5 is two. 00:20:04.930 --> 00:20:07.330 align:middle line:84% So that's consistent, hopefully, with what 00:20:07.330 --> 00:20:09.880 align:middle line:84% you would have expected for perfect covariance. 00:20:09.880 --> 00:20:11.560 align:middle line:84% You're getting the full information 00:20:11.560 --> 00:20:15.250 align:middle line:84% content, the full range of two bits, 00:20:15.250 --> 00:20:16.610 align:middle line:90% and so that's what we achieved. 00:20:16.610 --> 00:20:22.030 align:middle line:84% So now, as a controller is just further gratification 00:20:22.030 --> 00:20:24.700 align:middle line:84% that we actually understand this, 00:20:24.700 --> 00:20:27.250 align:middle line:84% as we're working through the example of comparing 00:20:27.250 --> 00:20:31.450 align:middle line:84% I equals 1 with J equals 2, so the two far columns. 00:20:31.450 --> 00:20:33.970 align:middle line:84% And here, you're familiar with I of 1. 00:20:33.970 --> 00:20:37.330 align:middle line:84% J equals 2 is always C, and so it's not 00:20:37.330 --> 00:20:42.650 align:middle line:84% covarying with the first column as in the previous example. 00:20:42.650 --> 00:20:46.460 align:middle line:84% So let's just work through it the same way. 00:20:46.460 --> 00:20:51.250 align:middle line:84% So the first term in the series is 0.25 again 00:20:51.250 --> 00:20:59.840 align:middle line:84% because the AC pair, not base pair, but pair of bases, 00:20:59.840 --> 00:21:02.560 align:middle line:84% occurs only once in the four of the multi-sequence alignment 00:21:02.560 --> 00:21:04.040 align:middle line:90% so that's 0.25. 00:21:04.040 --> 00:21:07.520 align:middle line:84% Then you have the logarithm base two of that same 0.25 00:21:07.520 --> 00:21:11.930 align:middle line:84% now normalized to the frequency of the A 00:21:11.930 --> 00:21:16.100 align:middle line:84% in its column, the one column, which is 0.25. 00:21:16.100 --> 00:21:19.940 align:middle line:84% And the C in the J equals 2 column, which is-- 00:21:19.940 --> 00:21:22.370 align:middle line:84% it's always there as unity, so it's one. 00:21:22.370 --> 00:21:24.200 align:middle line:84% So now that's the big change here, 00:21:24.200 --> 00:21:27.350 align:middle line:84% is instead of having 0.25 in both of the denominator terms, 00:21:27.350 --> 00:21:29.240 align:middle line:90% it's now 0.25 times one. 00:21:29.240 --> 00:21:33.950 align:middle line:84% It's now it's 0.25 over 0.25, so you have the log base 00:21:33.950 --> 00:21:36.680 align:middle line:90% 2 of one, which will be zero. 00:21:36.680 --> 00:21:38.400 align:middle line:84% And that zeros out the whole term. 00:21:38.400 --> 00:21:41.570 align:middle line:84% And so you have mutual information of zero, 00:21:41.570 --> 00:21:44.810 align:middle line:84% as you would expect from this particular toy example 00:21:44.810 --> 00:21:48.558 align:middle line:84% where columns one and two do not covary. 00:21:48.558 --> 00:21:51.225 align:middle line:84% And so this is the same formulas that was in the previous slide. 00:21:51.225 --> 00:21:55.380 align:middle line:84% And a generalization of the one that we walked through term 00:21:55.380 --> 00:21:56.530 align:middle line:90% by term. 00:21:56.530 --> 00:22:03.120 align:middle line:84% And here's the reference for that. 00:22:03.120 --> 00:22:06.800 align:middle line:84% So now how do we go-- so that was-- we've taken now 00:22:06.800 --> 00:22:09.530 align:middle line:84% hundreds, possibly thousands, of transfer RNAs. 00:22:09.530 --> 00:22:11.180 align:middle line:84% We've done a multi-sequence alignment. 00:22:11.180 --> 00:22:15.620 align:middle line:84% We've produced that mutual information pattern 00:22:15.620 --> 00:22:18.170 align:middle line:84% that we saw before with the one by 72-- 00:22:18.170 --> 00:22:21.830 align:middle line:84% one to 72 by one to 72 comparison 00:22:21.830 --> 00:22:26.810 align:middle line:84% where you got the spikes at each of the double helices. 00:22:26.810 --> 00:22:31.670 align:middle line:84% Now how do we turn that into more general practice. 00:22:31.670 --> 00:22:38.497 align:middle line:84% How do we generate secondary structures 00:22:38.497 --> 00:22:40.080 align:middle line:84% which are kind of at this intermediate 00:22:40.080 --> 00:22:42.455 align:middle line:84% in between the primary sequence and the three-dimensional 00:22:42.455 --> 00:22:49.530 align:middle line:84% structure using a particular class of experimental data 00:22:49.530 --> 00:22:52.290 align:middle line:90% combined with the sequence data. 00:22:52.290 --> 00:22:54.660 align:middle line:84% This does not necessarily require 00:22:54.660 --> 00:23:00.690 align:middle line:84% the large set of line sequences, but it obviously 00:23:00.690 --> 00:23:01.500 align:middle line:90% benefits from it. 00:23:01.500 --> 00:23:03.570 align:middle line:84% You could do a secondary structure 00:23:03.570 --> 00:23:08.550 align:middle line:84% for each element in the aligned sequence in order to-- 00:23:08.550 --> 00:23:12.900 align:middle line:84% and use the mutual information, if you have it. 00:23:12.900 --> 00:23:16.800 align:middle line:84% But let's just talk about just the simple application 00:23:16.800 --> 00:23:19.320 align:middle line:84% of these thermodynamic parameters to the prediction 00:23:19.320 --> 00:23:20.370 align:middle line:90% of secondary structure. 00:23:20.370 --> 00:23:22.110 align:middle line:84% And what are our expectations before we 00:23:22.110 --> 00:23:23.160 align:middle line:90% go through the algorithm. 00:23:23.160 --> 00:23:25.080 align:middle line:90% How good is it? 00:23:25.080 --> 00:23:30.810 align:middle line:84% In this fairly close to state-of-the-art paper, 00:23:30.810 --> 00:23:36.395 align:middle line:84% looking through over 700 generated structures, 00:23:36.395 --> 00:23:36.895 align:middle line:90% they have-- 00:23:36.895 --> 00:23:39.830 align:middle line:90% 00:23:39.830 --> 00:23:43.580 align:middle line:84% in each set, it contains one structure 00:23:43.580 --> 00:23:48.670 align:middle line:84% that, on average, has 86% of its known base pairs. 00:23:48.670 --> 00:23:50.580 align:middle line:84% That's not saying that it's necessarily 00:23:50.580 --> 00:23:54.270 align:middle line:84% identified as the top, the best, structure. 00:23:54.270 --> 00:23:56.850 align:middle line:84% It's saying that it has one structure by the criterion 00:23:56.850 --> 00:23:57.810 align:middle line:90% that they're using. 00:23:57.810 --> 00:24:00.450 align:middle line:90% This as a weak self praise. 00:24:00.450 --> 00:24:03.350 align:middle line:90% 00:24:03.350 --> 00:24:05.600 align:middle line:84% But let's walk through how that works. 00:24:05.600 --> 00:24:08.330 align:middle line:84% When someone says that they're going 00:24:08.330 --> 00:24:13.900 align:middle line:84% to predict a secondary structure or a three-dimensional 00:24:13.900 --> 00:24:16.800 align:middle line:84% structure from a primary sequence, 00:24:16.800 --> 00:24:20.760 align:middle line:84% more or less, from scratch, they really typically 00:24:20.760 --> 00:24:25.170 align:middle line:84% mean that there's going to be a variety of other chemical data 00:24:25.170 --> 00:24:28.260 align:middle line:84% that they take into account, but it will be generic data. 00:24:28.260 --> 00:24:30.390 align:middle line:84% It will not be specific chemical data 00:24:30.390 --> 00:24:31.920 align:middle line:90% for this particular molecule. 00:24:31.920 --> 00:24:34.050 align:middle line:84% And the generic data, in this case, 00:24:34.050 --> 00:24:40.730 align:middle line:84% are measurements of the thermodynamics of melting 00:24:40.730 --> 00:24:45.080 align:middle line:84% of model oligonucleotides, usually 00:24:45.080 --> 00:24:48.260 align:middle line:84% large amounts of them, monitored spectral photometrically. 00:24:48.260 --> 00:24:53.870 align:middle line:84% And from the temperatures of melting, basically, 00:24:53.870 --> 00:24:58.610 align:middle line:84% at equilibrium where you're getting half melted structures, 00:24:58.610 --> 00:25:01.490 align:middle line:84% you can determine the free energies 00:25:01.490 --> 00:25:04.100 align:middle line:84% where the negative free energies are the desirable ones, 00:25:04.100 --> 00:25:07.220 align:middle line:84% the ones that are likely to happen if you let 00:25:07.220 --> 00:25:10.200 align:middle line:90% the system go to equilibrium. 00:25:10.200 --> 00:25:14.030 align:middle line:84% And this is a kind of interesting application 00:25:14.030 --> 00:25:16.790 align:middle line:84% of the free energies for nucleic acids. 00:25:16.790 --> 00:25:21.140 align:middle line:84% Here, the algorithm that one uses 00:25:21.140 --> 00:25:26.370 align:middle line:84% is concerned mostly with adjacent base 00:25:26.370 --> 00:25:27.940 align:middle line:90% pairs of base pairs. 00:25:27.940 --> 00:25:31.380 align:middle line:84% So it's not a base pair, as you might 00:25:31.380 --> 00:25:36.150 align:middle line:84% think that the hydrogen bonds that determine the Watson-Crick 00:25:36.150 --> 00:25:38.820 align:middle line:84% and non Watson-Crick base pairs would dominate. 00:25:38.820 --> 00:25:41.310 align:middle line:84% Instead, it's the stacking interactions that dominate. 00:25:41.310 --> 00:25:43.680 align:middle line:84% And since it's the stacking interactions that dominate, 00:25:43.680 --> 00:25:46.740 align:middle line:84% the hydrogen bonds are basically exchanging a water hydrogen 00:25:46.740 --> 00:25:49.330 align:middle line:84% bond for a base pair hydrogen bond. 00:25:49.330 --> 00:25:51.820 align:middle line:84% It looks very specific but in terms of free energy, 00:25:51.820 --> 00:25:52.810 align:middle line:90% it's fairly weak. 00:25:52.810 --> 00:25:56.020 align:middle line:84% The free energy is determined more by the stacking 00:25:56.020 --> 00:25:59.080 align:middle line:84% of pi orbitals as depending on the geometry 00:25:59.080 --> 00:26:01.240 align:middle line:84% that you get, say, when you have a GC-- 00:26:01.240 --> 00:26:03.640 align:middle line:84% a CG base on top of an AU base pair 00:26:03.640 --> 00:26:05.920 align:middle line:84% here at the bottom of this helix. 00:26:05.920 --> 00:26:11.940 align:middle line:84% And that stack, it gives you a -2.1 kilocalories per mole. 00:26:11.940 --> 00:26:14.430 align:middle line:84% All the units on this are kilocalories per mole. 00:26:14.430 --> 00:26:18.090 align:middle line:84% And by going along and taking each of these stacks 00:26:18.090 --> 00:26:20.640 align:middle line:84% a pair of base pairs is what you're measuring. 00:26:20.640 --> 00:26:24.520 align:middle line:84% You can get all the negative free energy 00:26:24.520 --> 00:26:25.680 align:middle line:90% so they're stacking. 00:26:25.680 --> 00:26:27.930 align:middle line:84% Then you have some penalties, some things 00:26:27.930 --> 00:26:31.020 align:middle line:84% that are less favorable, that would not happen spontaneously 00:26:31.020 --> 00:26:34.320 align:middle line:84% if they did not have these mitigating negative 00:26:34.320 --> 00:26:36.330 align:middle line:84% free energies already accumulated, which would 00:26:36.330 --> 00:26:39.630 align:middle line:90% be the loop and the bulge here. 00:26:39.630 --> 00:26:42.450 align:middle line:84% The base pairs on either side of that bulge 00:26:42.450 --> 00:26:43.960 align:middle line:90% will stack up on one another. 00:26:43.960 --> 00:26:48.660 align:middle line:84% And that bulge will kind of flip out of the, otherwise, 00:26:48.660 --> 00:26:50.130 align:middle line:90% regular double helix. 00:26:50.130 --> 00:26:53.175 align:middle line:84% Similarly, bases at the end have a slight penalty. 00:26:53.175 --> 00:26:55.920 align:middle line:90% 00:26:55.920 --> 00:26:57.330 align:middle line:84% And so then you can add it all up 00:26:57.330 --> 00:26:58.913 align:middle line:84% and you can calculate it overall delta 00:26:58.913 --> 00:27:00.558 align:middle line:90% G for the entire structure. 00:27:00.558 --> 00:27:02.100 align:middle line:84% And if you do enough of these things, 00:27:02.100 --> 00:27:03.600 align:middle line:84% you can get a feeling for which ones 00:27:03.600 --> 00:27:07.845 align:middle line:84% are likely to be occurring in your RNAs. 00:27:07.845 --> 00:27:10.600 align:middle line:90% 00:27:10.600 --> 00:27:14.560 align:middle line:84% Now this should trigger in your mind, 00:27:14.560 --> 00:27:24.720 align:middle line:84% as the third example we've had where the conceit of a motif 00:27:24.720 --> 00:27:29.610 align:middle line:84% analysis, that you can do a multi-sequence alignment 00:27:29.610 --> 00:27:32.430 align:middle line:84% and each column and multi-sequence alignment 00:27:32.430 --> 00:27:33.250 align:middle line:90% is independent. 00:27:33.250 --> 00:27:36.120 align:middle line:84% This is the third example where that's not true. 00:27:36.120 --> 00:27:42.000 align:middle line:84% We have here the three energies are dependent on pairs 00:27:42.000 --> 00:27:43.560 align:middle line:90% of base pairs. 00:27:43.560 --> 00:27:49.980 align:middle line:84% The previous examples were the very distant connections 00:27:49.980 --> 00:27:52.440 align:middle line:84% that you can get in folding up a transfer RNA. 00:27:52.440 --> 00:27:57.460 align:middle line:84% And the earlier example was the CG dinucleotides. 00:27:57.460 --> 00:27:59.532 align:middle line:84% The assumption of independence of columns 00:27:59.532 --> 00:28:01.740 align:middle line:84% in a multi-sequence alignment is a very powerful one. 00:28:01.740 --> 00:28:03.323 align:middle line:84% I don't want to undermine it too much. 00:28:03.323 --> 00:28:06.600 align:middle line:84% But it doesn't hurt you to have three examples this early 00:28:06.600 --> 00:28:09.100 align:middle line:90% on in the course. 00:28:09.100 --> 00:28:11.695 align:middle line:84% Question the independence of columns 00:28:11.695 --> 00:28:15.020 align:middle line:84% in multi-sequence alignments-- very important thing 00:28:15.020 --> 00:28:15.520 align:middle line:90% to question. 00:28:15.520 --> 00:28:17.862 align:middle line:84% We've got mutual information theory 00:28:17.862 --> 00:28:19.570 align:middle line:84% that we had a couple of slides ago as one 00:28:19.570 --> 00:28:21.653 align:middle line:84% of the most powerful ways of questioning that when 00:28:21.653 --> 00:28:23.430 align:middle line:90% you see it. 00:28:23.430 --> 00:28:32.950 align:middle line:84% Now, that's the way that this particular base 00:28:32.950 --> 00:28:36.020 align:middle line:84% pairing, that we see here, is one example-- 00:28:36.020 --> 00:28:43.720 align:middle line:84% now you could take each of these and shift the right-hand half 00:28:43.720 --> 00:28:46.720 align:middle line:84% of the molecule relative to the left-hand by one base pair. 00:28:46.720 --> 00:28:51.080 align:middle line:84% That would give you a much poorer set 00:28:51.080 --> 00:28:57.200 align:middle line:84% of energies and much, much more bulges and more-- 00:28:57.200 --> 00:28:58.700 align:middle line:90% longer loops and so forth. 00:28:58.700 --> 00:29:02.090 align:middle line:84% And you end up with a poor delta G. 00:29:02.090 --> 00:29:04.400 align:middle line:84% And what you can do is you can rank 00:29:04.400 --> 00:29:11.090 align:middle line:84% and do one of these maximum-value searches 00:29:11.090 --> 00:29:12.530 align:middle line:90% by going through that. 00:29:12.530 --> 00:29:14.990 align:middle line:84% This should trigger in your mind this 00:29:14.990 --> 00:29:18.170 align:middle line:84% is another way of thinking about that search. 00:29:18.170 --> 00:29:22.670 align:middle line:84% You take the entire sequence, whether it's transfer RNA, 00:29:22.670 --> 00:29:26.570 align:middle line:84% or in this case, a 400 nucleic acid sequence, 00:29:26.570 --> 00:29:31.610 align:middle line:84% and you draw lines between every base where you 00:29:31.610 --> 00:29:33.380 align:middle line:90% have a favorable free energy. 00:29:33.380 --> 00:29:36.920 align:middle line:84% And you look for a set of lines which do not 00:29:36.920 --> 00:29:42.960 align:middle line:84% overlap one another, because these 00:29:42.960 --> 00:29:45.780 align:middle line:84% would represent short sequences of local-- 00:29:45.780 --> 00:29:48.000 align:middle line:84% you can think of this as a local sequence 00:29:48.000 --> 00:29:53.730 align:middle line:84% alignment between one half of the nucleic acid 00:29:53.730 --> 00:29:55.140 align:middle line:90% and the other half. 00:29:55.140 --> 00:29:57.960 align:middle line:84% Now this is not a sequence identity, remember, 00:29:57.960 --> 00:29:59.640 align:middle line:84% this is a sequence complementarity. 00:29:59.640 --> 00:30:02.400 align:middle line:84% That is to say, a reverse complement 00:30:02.400 --> 00:30:07.380 align:middle line:84% where complement means you've substituted As for Us and Cs 00:30:07.380 --> 00:30:08.310 align:middle line:90% for Gs. 00:30:08.310 --> 00:30:10.720 align:middle line:90% So you're looking for-- 00:30:10.720 --> 00:30:13.362 align:middle line:84% but in many other ways, this is analogous 00:30:13.362 --> 00:30:14.820 align:middle line:84% to the dynamic programming where we 00:30:14.820 --> 00:30:17.520 align:middle line:84% took two independent sequences and slid them along one 00:30:17.520 --> 00:30:20.700 align:middle line:84% another and allowed for insertions and deletions. 00:30:20.700 --> 00:30:23.100 align:middle line:84% In that-- in the dynamic equilibrium before, 00:30:23.100 --> 00:30:26.220 align:middle line:84% we did that formally, all possible such slippages 00:30:26.220 --> 00:30:30.820 align:middle line:84% by setting them as the two axes of a table 00:30:30.820 --> 00:30:34.860 align:middle line:84% and then filled in the squares for all the matches. 00:30:34.860 --> 00:30:38.760 align:middle line:84% Here, we would fill in the squares, not for the matches, 00:30:38.760 --> 00:30:42.510 align:middle line:84% but for the free energy of the stacking 00:30:42.510 --> 00:30:45.120 align:middle line:90% for these short subsequences. 00:30:45.120 --> 00:30:48.120 align:middle line:84% Now the reason that they don't cross over, 00:30:48.120 --> 00:30:51.540 align:middle line:84% and the reason for this little note in the lower 00:30:51.540 --> 00:30:53.760 align:middle line:84% left-hand corner of slide 11, that 00:30:53.760 --> 00:30:55.470 align:middle line:90% does not handle pseudo knots. 00:30:55.470 --> 00:30:57.990 align:middle line:84% Psuedo knots we'll explain in the next-- 00:30:57.990 --> 00:31:00.120 align:middle line:84% we'll show graphic examples in the next slide. 00:31:00.120 --> 00:31:01.770 align:middle line:84% But it basically means that if you 00:31:01.770 --> 00:31:04.560 align:middle line:84% allow such sequences to occur willy-nilly 00:31:04.560 --> 00:31:09.660 align:middle line:84% throughout the sequence, then you'll get these tangles 00:31:09.660 --> 00:31:13.522 align:middle line:84% that for a while people weren't sure 00:31:13.522 --> 00:31:14.730 align:middle line:90% whether they occurred or not. 00:31:14.730 --> 00:31:17.730 align:middle line:84% The one or two non Watson-Crick base pairs 00:31:17.730 --> 00:31:21.270 align:middle line:84% that you might find connecting up tRNA in these tangles 00:31:21.270 --> 00:31:23.610 align:middle line:84% were not considered long stretches 00:31:23.610 --> 00:31:25.380 align:middle line:90% that would connect loops. 00:31:25.380 --> 00:31:26.880 align:middle line:84% But since then, they've been proven 00:31:26.880 --> 00:31:29.130 align:middle line:84% to be of great biological significance. 00:31:29.130 --> 00:31:31.560 align:middle line:84% In any case, to do this without the pseudo knots, 00:31:31.560 --> 00:31:35.250 align:middle line:84% without allowing any crosses is still a challenging problem. 00:31:35.250 --> 00:31:38.370 align:middle line:84% It's basically the dynamic programming 00:31:38.370 --> 00:31:43.080 align:middle line:84% where N is the length of the primary sequence then takes 00:31:43.080 --> 00:31:46.980 align:middle line:84% on the order of N squared and compute time and space in order 00:31:46.980 --> 00:31:56.610 align:middle line:84% to figure out all the possible pairings that can occur. 00:31:56.610 --> 00:31:58.760 align:middle line:84% And then you go through and you rank them, 00:31:58.760 --> 00:32:00.440 align:middle line:84% which one gives the best free energy, 00:32:00.440 --> 00:32:04.040 align:middle line:84% and then you do the trace back and you get the top scores 00:32:04.040 --> 00:32:06.610 align:middle line:90% for that molecule. 00:32:06.610 --> 00:32:08.440 align:middle line:84% Now let's talk about pseudo notes. 00:32:08.440 --> 00:32:12.700 align:middle line:84% We excluded that, but now we'll reinvite it. 00:32:12.700 --> 00:32:17.980 align:middle line:84% We had those little ones, a couple of base pairs 00:32:17.980 --> 00:32:18.940 align:middle line:90% in the transfer RNA. 00:32:18.940 --> 00:32:22.530 align:middle line:84% But a much more dramatic one we alluded to in the second 00:32:22.530 --> 00:32:23.030 align:middle line:90% lecture. 00:32:23.030 --> 00:32:24.640 align:middle line:84% We talked about the genetic code. 00:32:24.640 --> 00:32:29.020 align:middle line:84% And in order to introduce you to exceptions to the genetic code, 00:32:29.020 --> 00:32:31.900 align:middle line:84% I gave an example where the ribosome 00:32:31.900 --> 00:32:34.120 align:middle line:84% jumped over 50 base pairs if presented 00:32:34.120 --> 00:32:35.470 align:middle line:90% with the right context. 00:32:35.470 --> 00:32:38.350 align:middle line:84% It didn't follow the normal code of having 00:32:38.350 --> 00:32:40.270 align:middle line:84% a triplet and another triplet right 00:32:40.270 --> 00:32:42.190 align:middle line:90% in a row with no punctuation. 00:32:42.190 --> 00:32:45.970 align:middle line:84% Here we had the punctuation that required, 00:32:45.970 --> 00:32:48.310 align:middle line:84% what may have slipped by at the time, a pseudo knot. 00:32:48.310 --> 00:32:52.630 align:middle line:84% And this is an example of once such pseudo knot in the best 00:32:52.630 --> 00:32:56.290 align:middle line:84% that we can do of a two dimensional schematic. 00:32:56.290 --> 00:32:58.510 align:middle line:84% And then something slightly better where 00:32:58.510 --> 00:33:02.020 align:middle line:84% we have a more three-dimensional and another three-dimensional 00:33:02.020 --> 00:33:02.770 align:middle line:90% view of this. 00:33:02.770 --> 00:33:05.470 align:middle line:84% And this is the RNA pseudo knot, which 00:33:05.470 --> 00:33:07.630 align:middle line:84% is-- one of them, which is responsible for frame 00:33:07.630 --> 00:33:11.060 align:middle line:84% shifting in the-- that breaks this genetic code. 00:33:11.060 --> 00:33:13.360 align:middle line:84% And so let's just follow how this goes. 00:33:13.360 --> 00:33:15.610 align:middle line:84% You've got basically a normal helix here 00:33:15.610 --> 00:33:18.390 align:middle line:84% at the bottom starting at the five prime in position one 00:33:18.390 --> 00:33:19.750 align:middle line:90% to seven. 00:33:19.750 --> 00:33:23.380 align:middle line:84% It would go through a normal five-base-- sorry-- 00:33:23.380 --> 00:33:26.200 align:middle line:84% six-base loop, from eight to 13, and then 00:33:26.200 --> 00:33:29.110 align:middle line:90% finish the stem 14 through 18. 00:33:29.110 --> 00:33:33.490 align:middle line:84% That would be a normal stem loop where the loop is six long, 00:33:33.490 --> 00:33:34.690 align:middle line:90% eight to 13. 00:33:34.690 --> 00:33:37.450 align:middle line:84% But at the end of eighteen, you have this little green loop 00:33:37.450 --> 00:33:41.830 align:middle line:84% that goes back and now makes a nice perfect four base 00:33:41.830 --> 00:33:48.410 align:middle line:90% Watson-Crick stem. 00:33:48.410 --> 00:33:50.970 align:middle line:84% Now in the middle of what would have been a loop-- 00:33:50.970 --> 00:33:54.980 align:middle line:84% and so this fold back is what we meant by a pseudo knot, 00:33:54.980 --> 00:33:56.480 align:middle line:84% and what would have been represented 00:33:56.480 --> 00:33:59.420 align:middle line:84% by a crossover of those red lines in the previous slide-- 00:33:59.420 --> 00:34:02.580 align:middle line:84% something that makes it much harder to compute. 00:34:02.580 --> 00:34:05.990 align:middle line:84% In fact, it makes it so much harder, in the next slide, 00:34:05.990 --> 00:34:09.230 align:middle line:84% that it goes up from an order of N squared, which 00:34:09.230 --> 00:34:11.900 align:middle line:84% is your typical dynamic programming 00:34:11.900 --> 00:34:18.000 align:middle line:84% in a pairwise alignment to order in sixth in CPU time and order 00:34:18.000 --> 00:34:20.882 align:middle line:84% into the fourth power in memory space 00:34:20.882 --> 00:34:22.590 align:middle line:84% that you need to set aside for storing up 00:34:22.590 --> 00:34:25.440 align:middle line:84% the table of possible pseudo knots 00:34:25.440 --> 00:34:27.750 align:middle line:84% that can occur in the context of the otherwise 00:34:27.750 --> 00:34:32.760 align:middle line:84% normal circle with the non-overlapping connections. 00:34:32.760 --> 00:34:41.610 align:middle line:84% This is a relatively recent innovation where a dynamic-- 00:34:41.610 --> 00:34:43.500 align:middle line:84% it's still a dynamic programming algorithm, 00:34:43.500 --> 00:34:47.699 align:middle line:84% it just has more possibilities, more complex-- 00:34:47.699 --> 00:34:50.670 align:middle line:90% higher algorithmic complexity. 00:34:50.670 --> 00:34:53.130 align:middle line:84% And the combination of the biological discovery 00:34:53.130 --> 00:34:56.010 align:middle line:84% to pseudo knots are important for frame shifting a variety 00:34:56.010 --> 00:34:57.930 align:middle line:90% of other biological phenomenon. 00:34:57.930 --> 00:35:02.130 align:middle line:84% And the now three-dimensional structure and now an algorithm 00:35:02.130 --> 00:35:05.160 align:middle line:84% puts pseudo knots well within the sort of things 00:35:05.160 --> 00:35:07.820 align:middle line:84% that you should feel comfortable. 00:35:07.820 --> 00:35:10.060 align:middle line:84% Now we're going to go back to hidden Markov 00:35:10.060 --> 00:35:13.480 align:middle line:84% models in a slightly more complicated context 00:35:13.480 --> 00:35:16.040 align:middle line:84% here, that we take the simplest one we could, 00:35:16.040 --> 00:35:22.740 align:middle line:84% which was a dinucleotide, simple, unfolded straight DNA. 00:35:22.740 --> 00:35:25.810 align:middle line:84% And the part that was hidden, as you will recall, 00:35:25.810 --> 00:35:29.285 align:middle line:84% was whether the the CG dinucleotides-- or sorry-- 00:35:29.285 --> 00:35:30.660 align:middle line:84% the dinucleotides, which could be 00:35:30.660 --> 00:35:34.650 align:middle line:84% any of the possible dinucleotides, 00:35:34.650 --> 00:35:38.880 align:middle line:84% including AA, CG, and so on, whether it was present in a CG 00:35:38.880 --> 00:35:41.800 align:middle line:84% island, or whether it was in a region of the chromosome 00:35:41.800 --> 00:35:43.830 align:middle line:84% which was likely to have CGs, or whether it 00:35:43.830 --> 00:35:51.060 align:middle line:84% was in a CG ocean which was low in CG dinucleotide content. 00:35:51.060 --> 00:35:53.160 align:middle line:84% So the hidden part was the plus-minus 00:35:53.160 --> 00:35:55.290 align:middle line:84% whether it was in an island or not. 00:35:55.290 --> 00:35:58.710 align:middle line:84% Now what we're going to do now is take this and transfer this 00:35:58.710 --> 00:36:03.660 align:middle line:84% over to the kinds of motifs we are finding in RNAs, 00:36:03.660 --> 00:36:06.240 align:middle line:84% like transfer RNA and another class of RNA, 00:36:06.240 --> 00:36:11.160 align:middle line:84% and say OK, now, whether it's the hidden part of the Markov 00:36:11.160 --> 00:36:12.865 align:middle line:84% model is these transition probabilities. 00:36:12.865 --> 00:36:15.240 align:middle line:84% The hidden part is whether it's in a particular secondary 00:36:15.240 --> 00:36:18.030 align:middle line:84% structure or not, not whether it's in an island or not, 00:36:18.030 --> 00:36:20.650 align:middle line:90% but a secondary structure. 00:36:20.650 --> 00:36:22.492 align:middle line:84% In this particular case we're going 00:36:22.492 --> 00:36:23.950 align:middle line:84% to talk about is a very interesting 00:36:23.950 --> 00:36:33.150 align:middle line:84% biological illustration where the hidden Markov models will 00:36:33.150 --> 00:36:36.720 align:middle line:84% be modeling these boxes, these motifs, that are involved 00:36:36.720 --> 00:36:41.650 align:middle line:84% in base pairing or recognition that 00:36:41.650 --> 00:36:44.486 align:middle line:84% forms the secondary structure of-- 00:36:44.486 --> 00:36:48.300 align:middle line:84% if it's necessary for guiding a particular enzyme. 00:36:48.300 --> 00:36:52.380 align:middle line:84% Now, remember we had all these modified bases that we used, 00:36:52.380 --> 00:36:53.850 align:middle line:90% that we saw in transfer RNA. 00:36:53.850 --> 00:36:56.160 align:middle line:84% Some of those are simple protein interactions 00:36:56.160 --> 00:36:58.710 align:middle line:84% with the transfer RNA that adds a methyl group here or there. 00:36:58.710 --> 00:37:01.560 align:middle line:84% It turns out that all of the methyl groups, the O2 00:37:01.560 --> 00:37:07.850 align:middle line:84% prime methyl groups, these are on the sugar of the ribose. 00:37:07.850 --> 00:37:13.970 align:middle line:84% In ribosomal RNA, a few, just a small number of a few dozen 00:37:13.970 --> 00:37:20.150 align:middle line:84% of the ribosomes in this multikilobase ribosomal RNA 00:37:20.150 --> 00:37:21.890 align:middle line:90% are methylated O2 position. 00:37:21.890 --> 00:37:23.975 align:middle line:84% How does the enzyme know, or the enzymes, 00:37:23.975 --> 00:37:26.540 align:middle line:90% know to get exactly those bases? 00:37:26.540 --> 00:37:35.470 align:middle line:84% The way it knows is it doesn't use pure protein brute force 00:37:35.470 --> 00:37:38.710 align:middle line:84% to make a complimentary surface of protein nucleic acid. 00:37:38.710 --> 00:37:43.890 align:middle line:84% It actually uses this elegance of base pairing 00:37:43.890 --> 00:37:45.420 align:middle line:90% to make a guide sequence. 00:37:45.420 --> 00:37:47.970 align:middle line:84% And so what it's looking for is-- 00:37:47.970 --> 00:37:53.070 align:middle line:84% the protein cooperates with a small RNA, so-called snow RNA, 00:37:53.070 --> 00:37:57.120 align:middle line:84% or small nucleolar RNA, to find a place 00:37:57.120 --> 00:37:59.550 align:middle line:84% where the snow will recognize the place that you 00:37:59.550 --> 00:38:00.390 align:middle line:90% want to methylate. 00:38:00.390 --> 00:38:03.990 align:middle line:84% And then the protein methylates the base 00:38:03.990 --> 00:38:06.360 align:middle line:84% in the middle of that guide sequence. 00:38:06.360 --> 00:38:14.140 align:middle line:84% So then the game, the computational biology game, 00:38:14.140 --> 00:38:17.260 align:middle line:84% that these authors played was how can we 00:38:17.260 --> 00:38:21.790 align:middle line:84% find all the small RNAs, the snow RNAs, 00:38:21.790 --> 00:38:29.380 align:middle line:84% present in a genome when we very little about that the genome? 00:38:29.380 --> 00:38:31.480 align:middle line:84% What they knew was they knew the genome sequence. 00:38:31.480 --> 00:38:32.530 align:middle line:90% This is for yeast. 00:38:32.530 --> 00:38:37.450 align:middle line:84% They had a few examples of snow RNAs in humans-- 00:38:37.450 --> 00:38:39.010 align:middle line:90% almost none in yeast. 00:38:39.010 --> 00:38:42.310 align:middle line:84% They had the subsequence of the ribosomal RNA, of course. 00:38:42.310 --> 00:38:47.140 align:middle line:84% And they could-- what they wanted 00:38:47.140 --> 00:38:51.460 align:middle line:84% to do then is ask where in the genome do we have little guide 00:38:51.460 --> 00:38:55.330 align:middle line:84% sequences flanked by some of these other motifs 00:38:55.330 --> 00:39:00.520 align:middle line:84% and characteristics like a base pair, 4-8 base pair stem, 00:39:00.520 --> 00:39:02.347 align:middle line:84% that will match the ribosomal RNA. 00:39:02.347 --> 00:39:04.180 align:middle line:84% So you basically march along the algorithms, 00:39:04.180 --> 00:39:06.520 align:middle line:84% you march along the ribosomal RNA looking for matches 00:39:06.520 --> 00:39:07.820 align:middle line:90% elsewhere in the genome. 00:39:07.820 --> 00:39:10.237 align:middle line:84% And then ask whether those matches elsewhere in the genome 00:39:10.237 --> 00:39:13.305 align:middle line:84% have some of these other contexts, features. 00:39:13.305 --> 00:39:15.805 align:middle line:84% You can see this is going to be a more complicated algorithm 00:39:15.805 --> 00:39:18.920 align:middle line:90% than just looking for CGs. 00:39:18.920 --> 00:39:22.360 align:middle line:90% So this is how it works. 00:39:22.360 --> 00:39:28.840 align:middle line:84% That stem that we had is now item number one. 00:39:28.840 --> 00:39:32.410 align:middle line:84% The various boxes which were basically sequences 00:39:32.410 --> 00:39:36.410 align:middle line:84% are now turned into ungapped hidden Markov models. 00:39:36.410 --> 00:39:39.610 align:middle line:84% The hidden part of it is whether it is present 00:39:39.610 --> 00:39:45.130 align:middle line:84% or not in the context that adds up to this guide sequence. 00:39:45.130 --> 00:39:48.520 align:middle line:84% The guide sequence itself is a hidden Markov model which 00:39:48.520 --> 00:39:53.990 align:middle line:84% has to be an imperfect, probably imperfect, duplex 00:39:53.990 --> 00:39:56.510 align:middle line:90% with the ribosomal RNA. 00:39:56.510 --> 00:39:59.810 align:middle line:90% So that's how that's modeled. 00:39:59.810 --> 00:40:01.910 align:middle line:84% The most complicated is that terminal 00:40:01.910 --> 00:40:04.700 align:middle line:84% stem number one, which is a so-called Stochastic 00:40:04.700 --> 00:40:05.930 align:middle line:90% Context-Free Grammar. 00:40:05.930 --> 00:40:08.630 align:middle line:90% That's what the SCFG stands for. 00:40:08.630 --> 00:40:11.570 align:middle line:84% And that just means that it is even less constrained 00:40:11.570 --> 00:40:12.470 align:middle line:90% than the HMM. 00:40:12.470 --> 00:40:15.170 align:middle line:84% The HMM is less constrained than a simple motif, 00:40:15.170 --> 00:40:17.930 align:middle line:84% which is less constrained than, say, a consensus sequence. 00:40:17.930 --> 00:40:21.500 align:middle line:84% It is constrained, it has the grammar, if you will, 00:40:21.500 --> 00:40:24.470 align:middle line:84% or the particular rules for the base pairing 00:40:24.470 --> 00:40:26.990 align:middle line:84% that have to occur over a certain region 00:40:26.990 --> 00:40:29.780 align:middle line:84% in a certain part of this putative snow RNA. 00:40:29.780 --> 00:40:31.700 align:middle line:84% So anyway, you apply each of these criteria 00:40:31.700 --> 00:40:33.810 align:middle line:84% and you have transition probabilities 00:40:33.810 --> 00:40:37.080 align:middle line:84% which come in from a learning set, such as the human snow 00:40:37.080 --> 00:40:37.610 align:middle line:90% RNAs. 00:40:37.610 --> 00:40:39.193 align:middle line:84% You have a learning set that tells you 00:40:39.193 --> 00:40:41.210 align:middle line:84% what these transition probabilities will be. 00:40:41.210 --> 00:40:43.970 align:middle line:84% And you and you now apply this to the entire yeast genome, 00:40:43.970 --> 00:40:46.610 align:middle line:84% and you get a bunch of candidates, snow RNA 00:40:46.610 --> 00:40:47.780 align:middle line:90% encoding genes. 00:40:47.780 --> 00:40:50.450 align:middle line:84% Now you can't use things like the long open reading 00:40:50.450 --> 00:40:52.800 align:middle line:84% frames that you normally would use for finding genes. 00:40:52.800 --> 00:40:54.800 align:middle line:90% So this is a very valuable tool. 00:40:54.800 --> 00:40:57.080 align:middle line:84% But now how do you convince yourself 00:40:57.080 --> 00:41:03.920 align:middle line:84% that this is a gene, that this actually encodes a snow RNA, 00:41:03.920 --> 00:41:07.160 align:middle line:84% and that those are responsible for guiding 00:41:07.160 --> 00:41:11.300 align:middle line:84% the methylation at particular positions in the ribosomal RNA. 00:41:11.300 --> 00:41:14.550 align:middle line:90% The way you do that is-- 00:41:14.550 --> 00:41:16.770 align:middle line:84% well, before we get to how you do that, 00:41:16.770 --> 00:41:20.190 align:middle line:84% we want to ask how does this algorithm perform 00:41:20.190 --> 00:41:21.540 align:middle line:90% relative to other-- 00:41:21.540 --> 00:41:23.130 align:middle line:84% the few other algorithms there are 00:41:23.130 --> 00:41:26.260 align:middle line:84% for finding genes which do not encode proteins. 00:41:26.260 --> 00:41:30.750 align:middle line:84% And the first of these actually dates well before 1991. 00:41:30.750 --> 00:41:33.510 align:middle line:84% But there were ways of looking for transfer RNAs in sequence. 00:41:33.510 --> 00:41:36.090 align:middle line:84% They would use everything we know about transfer RNAs-- 00:41:36.090 --> 00:41:38.130 align:middle line:84% the little boxes that are conserved 00:41:38.130 --> 00:41:41.340 align:middle line:84% as sequences, the regions that are conserved only-- 00:41:41.340 --> 00:41:45.660 align:middle line:84% not as sequences, but as base pairing potential, et cetera. 00:41:45.660 --> 00:41:48.090 align:middle line:90% The loop lengths look-- 00:41:48.090 --> 00:41:49.020 align:middle line:90% are constrained. 00:41:49.020 --> 00:41:52.650 align:middle line:84% All the constraints that you can muster back in '91 00:41:52.650 --> 00:41:53.580 align:middle line:90% were applied. 00:41:53.580 --> 00:41:55.800 align:middle line:90% And it was fairly slow. 00:41:55.800 --> 00:42:01.590 align:middle line:84% It would only do 400 base pairs of genome chunk per second. 00:42:01.590 --> 00:42:05.580 align:middle line:84% And when you have genomes on the order of many mega bases, 00:42:05.580 --> 00:42:07.050 align:middle line:90% this is slow. 00:42:07.050 --> 00:42:11.865 align:middle line:84% And it had-- it missed about 5% of the true positives. 00:42:11.865 --> 00:42:13.590 align:middle line:90% It had 95%. 00:42:13.590 --> 00:42:17.280 align:middle line:84% And false positives sounds impressive-- only 10 00:42:17.280 --> 00:42:18.180 align:middle line:90% to the minus 6. 00:42:18.180 --> 00:42:20.370 align:middle line:84% But when you think of a double stranded-- 00:42:20.370 --> 00:42:26.380 align:middle line:84% both strands of E. coli being about 10 million bases, 00:42:26.380 --> 00:42:28.750 align:middle line:84% then this is about four false positives. 00:42:28.750 --> 00:42:31.540 align:middle line:84% And bigger genomes, of course, would be even a larger 00:42:31.540 --> 00:42:33.160 align:middle line:90% number in an absolute scale. 00:42:33.160 --> 00:42:39.130 align:middle line:84% So then, six years later, the speed is now 100 times faster. 00:42:39.130 --> 00:42:42.370 align:middle line:84% You're now only missing 0.5% of the true positives instead 00:42:42.370 --> 00:42:43.420 align:middle line:90% of 5%. 00:42:43.420 --> 00:42:47.000 align:middle line:84% And the false positives is now vanishingly small. 00:42:47.000 --> 00:42:52.828 align:middle line:84% So very often you can just arbitrarily trade off 00:42:52.828 --> 00:42:54.370 align:middle line:84% the number of true positives you miss 00:42:54.370 --> 00:42:55.787 align:middle line:84% with the number of false positives 00:42:55.787 --> 00:42:57.490 align:middle line:90% you get and make one-- 00:42:57.490 --> 00:42:58.910 align:middle line:90% take one advantage of the other. 00:42:58.910 --> 00:43:00.550 align:middle line:84% But here, it was a win-win situation. 00:43:00.550 --> 00:43:03.110 align:middle line:84% They both went in a favorable direction. 00:43:03.110 --> 00:43:06.470 align:middle line:84% So how do the snow RNAs compare with that? 00:43:06.470 --> 00:43:08.190 align:middle line:90% Here, another two years passed. 00:43:08.190 --> 00:43:10.710 align:middle line:84% We have the snow RNAs are just starting out. 00:43:10.710 --> 00:43:15.170 align:middle line:84% They have, probably, a little better than 93% true positives. 00:43:15.170 --> 00:43:17.000 align:middle line:84% This is not as good as transfer RNAs. 00:43:17.000 --> 00:43:23.720 align:middle line:84% This may be-- this may improve or it may not. 00:43:23.720 --> 00:43:28.030 align:middle line:84% The false positive rate is acceptable. 00:43:28.030 --> 00:43:31.960 align:middle line:84% So then the question becomes, how do you track down-- 00:43:31.960 --> 00:43:33.340 align:middle line:84% after you track down these genes, 00:43:33.340 --> 00:43:37.390 align:middle line:84% how do you then prove that they do what you think they do, 00:43:37.390 --> 00:43:40.300 align:middle line:84% that they actually are responsible for methylating 00:43:40.300 --> 00:43:44.220 align:middle line:84% the riboses or the bases in question? 00:43:44.220 --> 00:43:50.960 align:middle line:84% So it turns out that the technology we set up 00:43:50.960 --> 00:43:53.150 align:middle line:84% in the sequencing and genotyping lecture 00:43:53.150 --> 00:43:59.510 align:middle line:84% where you extend with DNA polymerase a primer 00:43:59.510 --> 00:44:02.360 align:middle line:84% on a template so the primer binds to the template 00:44:02.360 --> 00:44:05.270 align:middle line:84% and you extend by either many base pairs, 00:44:05.270 --> 00:44:07.610 align:middle line:84% as in the conventional dideoxy sequencing, 00:44:07.610 --> 00:44:11.090 align:middle line:84% or one or two base pairs in some of the more 00:44:11.090 --> 00:44:16.200 align:middle line:84% up-and-coming genotyping methods. 00:44:16.200 --> 00:44:19.860 align:middle line:84% Those extension methods, those DNA polymerase-based extension 00:44:19.860 --> 00:44:26.160 align:middle line:84% methods will stall when you run into this particular kind 00:44:26.160 --> 00:44:29.880 align:middle line:84% of modified base where a bulky group is introduced 00:44:29.880 --> 00:44:34.217 align:middle line:84% onto the two prime position of the ribose on the template. 00:44:34.217 --> 00:44:36.550 align:middle line:84% So you're extending the primer, sitting on the template, 00:44:36.550 --> 00:44:37.890 align:middle line:90% and it will stall here. 00:44:37.890 --> 00:44:40.800 align:middle line:84% And it stalls more when you decrease the concentration 00:44:40.800 --> 00:44:44.910 align:middle line:84% of the deoxynucleotide triphosphates in the extension 00:44:44.910 --> 00:44:45.610 align:middle line:90% reaction. 00:44:45.610 --> 00:44:46.650 align:middle line:84% So that's what these little wedges 00:44:46.650 --> 00:44:48.510 align:middle line:84% mean at the top of each of these columns. 00:44:48.510 --> 00:44:53.310 align:middle line:84% They've done an extension with all four triphosphates present, 00:44:53.310 --> 00:44:55.110 align:middle line:84% either in high amounts at the big end 00:44:55.110 --> 00:44:57.600 align:middle line:84% of the wedge or low amounts in the small ends of the wedge. 00:44:57.600 --> 00:45:00.150 align:middle line:84% And to tell where you are in the sequence-- this 00:45:00.150 --> 00:45:01.680 align:middle line:84% is what using reverse transcriptase 00:45:01.680 --> 00:45:05.370 align:middle line:84% polymerase on a ribosomal RNA template-- 00:45:05.370 --> 00:45:06.990 align:middle line:84% to find out where you are, you do 00:45:06.990 --> 00:45:09.250 align:middle line:84% this dideoxy, which is basically the conventional DNA 00:45:09.250 --> 00:45:09.750 align:middle line:90% sequencing. 00:45:09.750 --> 00:45:12.055 align:middle line:84% Where you terminate, it either is used Us, 00:45:12.055 --> 00:45:13.305 align:middle line:90% Gs, Cs, or As in the template. 00:45:13.305 --> 00:45:15.932 align:middle line:90% 00:45:15.932 --> 00:45:17.265 align:middle line:90% This allows you to get oriented. 00:45:17.265 --> 00:45:20.370 align:middle line:84% And basically, your sequencing on the far left-hand 00:45:20.370 --> 00:45:21.390 align:middle line:90% set of lanes. 00:45:21.390 --> 00:45:25.440 align:middle line:84% And these pause sites our present, 00:45:25.440 --> 00:45:29.460 align:middle line:84% say, the wild-type is the first pair of lanes 00:45:29.460 --> 00:45:33.480 align:middle line:84% next to the sequence lanes on the left-hand side 00:45:33.480 --> 00:45:34.480 align:middle line:90% of this display. 00:45:34.480 --> 00:45:38.070 align:middle line:84% And you can see there's a pause at every single known 00:45:38.070 --> 00:45:39.300 align:middle line:90% methylated base. 00:45:39.300 --> 00:45:44.040 align:middle line:84% You can determine methylated bases by other methods as well. 00:45:44.040 --> 00:45:47.850 align:middle line:84% But so now then, the computational biology 00:45:47.850 --> 00:45:50.550 align:middle line:90% predicted a set of snow genes. 00:45:50.550 --> 00:45:53.820 align:middle line:84% In fact, ultimately, all of the snow genes in the yeast 00:45:53.820 --> 00:45:57.390 align:middle line:84% genome we think, explaining all of the methyl groups, at least. 00:45:57.390 --> 00:46:00.540 align:middle line:84% And one by one, these were knocked out cleanly 00:46:00.540 --> 00:46:03.648 align:middle line:84% so that there's no gene there anymore for the small RNA. 00:46:03.648 --> 00:46:05.190 align:middle line:84% And then you ask, well, how does this 00:46:05.190 --> 00:46:07.710 align:middle line:84% affect the methylation as detected by this extension 00:46:07.710 --> 00:46:08.340 align:middle line:90% assay? 00:46:08.340 --> 00:46:12.760 align:middle line:84% And if you look on the far right-hand side for deleted 00:46:12.760 --> 00:46:17.040 align:middle line:84% number 40, and you can see that position number 00:46:17.040 --> 00:46:21.330 align:middle line:84% 596, near the bottom, circled in red, which 00:46:21.330 --> 00:46:23.730 align:middle line:84% is present in the wild type and all the other mutants, 00:46:23.730 --> 00:46:26.680 align:middle line:84% is absent from that particular mutant number 40. 00:46:26.680 --> 00:46:28.350 align:middle line:90% So there's no pause there. 00:46:28.350 --> 00:46:30.480 align:middle line:84% One infers there's no methylation. 00:46:30.480 --> 00:46:34.050 align:middle line:84% And that was the specific site that that snow RNA guide 00:46:34.050 --> 00:46:36.810 align:middle line:90% sequence was predicted to bind. 00:46:36.810 --> 00:46:40.110 align:middle line:84% It's aligned with the position in the guide 00:46:40.110 --> 00:46:44.790 align:middle line:84% where you expect there to be a methylation occurring. 00:46:44.790 --> 00:46:47.700 align:middle line:84% And you can see in each lane there's a different circled 00:46:47.700 --> 00:46:51.780 align:middle line:90% red missing black pause site. 00:46:51.780 --> 00:46:57.490 align:middle line:84% And until we get to the one in the middle, 00:46:57.490 --> 00:47:00.250 align:middle line:84% Newton numbers for snow RNA number 60-- 00:47:00.250 --> 00:47:03.880 align:middle line:84% and here's actually two missing bands in the same lane. 00:47:03.880 --> 00:47:06.520 align:middle line:84% And how that can occur, there's two different ways 00:47:06.520 --> 00:47:08.920 align:middle line:90% that a snow RNA can-- 00:47:08.920 --> 00:47:11.200 align:middle line:84% knocking out a single gene, a single snow RNA 00:47:11.200 --> 00:47:14.120 align:middle line:84% can have an effect on two different methyl groups. 00:47:14.120 --> 00:47:21.340 align:middle line:84% One is if the guide sequence can bind to two different places 00:47:21.340 --> 00:47:22.510 align:middle line:90% in the ribosomal RNA. 00:47:22.510 --> 00:47:25.270 align:middle line:84% And the other is if there are two guides sequences 00:47:25.270 --> 00:47:26.620 align:middle line:90% within the same snow RNA. 00:47:26.620 --> 00:47:31.340 align:middle line:90% 00:47:31.340 --> 00:47:35.610 align:middle line:84% So now that we have at least some grounding 00:47:35.610 --> 00:47:40.200 align:middle line:84% in the kind of structures that can occur, now 00:47:40.200 --> 00:47:43.020 align:middle line:84% we're going to ask how we monitor 00:47:43.020 --> 00:47:47.010 align:middle line:84% and measure the amounts of these structures 00:47:47.010 --> 00:47:50.150 align:middle line:90% in biological systems. 00:47:50.150 --> 00:47:53.450 align:middle line:84% And we will also see how these structures impact the methods 00:47:53.450 --> 00:47:58.240 align:middle line:84% that we use for the quantitation of the structures. 00:47:58.240 --> 00:48:07.740 align:middle line:84% So we have choices of molecule that we're 00:48:07.740 --> 00:48:10.230 align:middle line:84% going to measure when we're monitoring 00:48:10.230 --> 00:48:12.510 align:middle line:84% the various molecules in the cell. 00:48:12.510 --> 00:48:14.160 align:middle line:90% Why are we focusing on RNA? 00:48:14.160 --> 00:48:18.390 align:middle line:84% Well, part of it is because of its nice structural continuity 00:48:18.390 --> 00:48:21.390 align:middle line:84% between the simple DNA and the very complicated proteins. 00:48:21.390 --> 00:48:25.470 align:middle line:84% But the other is that if we want to study different points 00:48:25.470 --> 00:48:28.812 align:middle line:84% in the regulatory and metabolic networks 00:48:28.812 --> 00:48:31.020 align:middle line:84% that we'll be talking about at the end of this course 00:48:31.020 --> 00:48:36.530 align:middle line:84% having to do with systems biology, if we choose-- 00:48:36.530 --> 00:48:40.310 align:middle line:84% we want-- every part of it is subject to some kind 00:48:40.310 --> 00:48:41.180 align:middle line:90% of control. 00:48:41.180 --> 00:48:44.240 align:middle line:84% Transcriptional control is one of the early stages. 00:48:44.240 --> 00:48:46.820 align:middle line:84% And then there are many stages subsequence to that lead 00:48:46.820 --> 00:48:50.480 align:middle line:84% to the protein and ultimate phenotypes that 00:48:50.480 --> 00:48:52.820 align:middle line:84% result in proliferation of the species. 00:48:52.820 --> 00:48:55.310 align:middle line:84% If you want to look at transcriptional control, 00:48:55.310 --> 00:48:59.360 align:middle line:84% it would not do well to study protein 00:48:59.360 --> 00:49:02.510 align:middle line:84% because the closest thing to transcriptional control 00:49:02.510 --> 00:49:05.840 align:middle line:84% that you can measure is the RNA products. 00:49:05.840 --> 00:49:08.540 align:middle line:84% You can study the transcriptional control itself 00:49:08.540 --> 00:49:11.210 align:middle line:84% directly, as well, by studying the DNA protein interactions. 00:49:11.210 --> 00:49:14.450 align:middle line:84% But if you want to measure, a defusible molecule RNA 00:49:14.450 --> 00:49:15.690 align:middle line:90% is the thing to do. 00:49:15.690 --> 00:49:18.770 align:middle line:84% And there are multiple different methods 00:49:18.770 --> 00:49:22.100 align:middle line:84% for getting at a co-regulated set of genes, co-regulated 00:49:22.100 --> 00:49:24.290 align:middle line:90% at the transcriptional level. 00:49:24.290 --> 00:49:26.990 align:middle line:90% We'll illustrate a few here. 00:49:26.990 --> 00:49:28.850 align:middle line:84% And why do we need multiple methods? 00:49:28.850 --> 00:49:31.820 align:middle line:84% Well, we've talked about random and systematic errors. 00:49:31.820 --> 00:49:37.230 align:middle line:84% Random errors you can compensate by repeating the experiment. 00:49:37.230 --> 00:49:39.307 align:middle line:84% The random errors will average out, ultimately. 00:49:39.307 --> 00:49:41.640 align:middle line:84% Systematic errors will happen the same way over and over 00:49:41.640 --> 00:49:41.980 align:middle line:90% again. 00:49:41.980 --> 00:49:43.813 align:middle line:84% So you want to have something out of the box 00:49:43.813 --> 00:49:47.190 align:middle line:84% to allow you to check it, or to model it, 00:49:47.190 --> 00:49:51.000 align:middle line:84% or to allow you to do integration, 00:49:51.000 --> 00:49:54.160 align:middle line:84% as you might want to have in complicated systems. 00:49:54.160 --> 00:49:57.180 align:middle line:84% So here, just to start us thinking about integration 00:49:57.180 --> 00:50:00.750 align:middle line:84% and checking different ways of getting transcriptional 00:50:00.750 --> 00:50:05.760 align:middle line:84% co-regulation, let's think about-- 00:50:05.760 --> 00:50:09.150 align:middle line:84% if you look through all the proteins that occur, 00:50:09.150 --> 00:50:13.080 align:middle line:84% you will find proteins that occur together, frequently, 00:50:13.080 --> 00:50:17.610 align:middle line:84% as either as fusions or as separate proteins. 00:50:17.610 --> 00:50:21.240 align:middle line:84% Where in operands, they'll occur as coding regions 00:50:21.240 --> 00:50:23.580 align:middle line:84% that are clustered together in some species 00:50:23.580 --> 00:50:27.350 align:middle line:84% or maybe less clustered in other species. 00:50:27.350 --> 00:50:31.840 align:middle line:84% When we have metabolic pathways where a small molecule will 00:50:31.840 --> 00:50:34.480 align:middle line:90% be shared by-- 00:50:34.480 --> 00:50:38.000 align:middle line:84% as the product of one will be the substrate of another, 00:50:38.000 --> 00:50:43.910 align:middle line:84% and so on, you'll have this chain of events as in the lower 00:50:43.910 --> 00:50:45.350 align:middle line:90% left-hand corner. 00:50:45.350 --> 00:50:54.250 align:middle line:84% And these sets of enzymes that need to be working together, 00:50:54.250 --> 00:50:55.420 align:middle line:90% need to be co-expressed. 00:50:55.420 --> 00:50:59.290 align:middle line:84% They need to come up together and go down together 00:50:59.290 --> 00:51:00.430 align:middle line:90% when they're not needed. 00:51:00.430 --> 00:51:02.620 align:middle line:84% They need to come up when they're suddenly needed. 00:51:02.620 --> 00:51:05.920 align:middle line:84% And so, you might have an entire pathway, or a set of pathways, 00:51:05.920 --> 00:51:07.120 align:middle line:90% that are co-expressed. 00:51:07.120 --> 00:51:11.200 align:middle line:84% And one way to do that is to cluster them in the genome. 00:51:11.200 --> 00:51:13.840 align:middle line:84% When they are co-expressed, you will sometimes 00:51:13.840 --> 00:51:18.250 align:middle line:84% find upstream of them, motifs, such as this. 00:51:18.250 --> 00:51:23.170 align:middle line:84% Again, here's the two bits for the vertical scale, 00:51:23.170 --> 00:51:26.650 align:middle line:90% where this might be enriched. 00:51:26.650 --> 00:51:29.020 align:middle line:84% And so this would be another indication-- so 00:51:29.020 --> 00:51:32.290 align:middle line:84% when you find these in front of genes, 00:51:32.290 --> 00:51:35.670 align:middle line:84% you might expect them to be co-regulated. 00:51:35.670 --> 00:51:39.990 align:middle line:84% When you find a set of proteins that are consistently together 00:51:39.990 --> 00:51:42.420 align:middle line:84% in different organisms, so-called phylogenetic 00:51:42.420 --> 00:51:47.490 align:middle line:84% profiles, you will find that this set of proteins that is 00:51:47.490 --> 00:51:51.150 align:middle line:84% involved in a common enzymatic pathway, metabolic pathway, 00:51:51.150 --> 00:51:56.105 align:middle line:84% are not only co-regulated and found together but in-- 00:51:56.105 --> 00:51:57.480 align:middle line:84% along the chromosome, but they're 00:51:57.480 --> 00:52:00.300 align:middle line:84% found together when you go through many different species. 00:52:00.300 --> 00:52:03.850 align:middle line:84% They will be deleted or inserted as a block, 00:52:03.850 --> 00:52:05.850 align:middle line:84% or they'll be found scattered around the genome. 00:52:05.850 --> 00:52:08.910 align:middle line:84% But you'll find that when one disappears, 00:52:08.910 --> 00:52:12.030 align:middle line:84% they all disappear, in general, statistically speaking. 00:52:12.030 --> 00:52:14.610 align:middle line:84% It's this phylogenetic co-occurrence 00:52:14.610 --> 00:52:16.900 align:middle line:84% is another clue that you might expect 00:52:16.900 --> 00:52:20.190 align:middle line:84% them to be co-regulated in those genomes in which there they 00:52:20.190 --> 00:52:22.180 align:middle line:90% do co-occur. 00:52:22.180 --> 00:52:25.030 align:middle line:84% Anyway, and microarrays will be-- 00:52:25.030 --> 00:52:27.190 align:middle line:84% and variations on that theme will be the main thing 00:52:27.190 --> 00:52:28.217 align:middle line:90% that we'll talk about. 00:52:28.217 --> 00:52:29.800 align:middle line:84% But I wanted to put it in the context. 00:52:29.800 --> 00:52:35.390 align:middle line:84% And I'll just expand on one of these at the bottom 00:52:35.390 --> 00:52:40.980 align:middle line:90% here in slide 22. 00:52:40.980 --> 00:52:44.070 align:middle line:84% This is an algorithm for reconstructing 00:52:44.070 --> 00:52:49.440 align:middle line:84% likely combinations where in some organisms 00:52:49.440 --> 00:52:52.650 align:middle line:84% you might have the entire biosynthetic pathway 00:52:52.650 --> 00:52:58.170 align:middle line:84% as a series of genes which encode one by one, all 00:52:58.170 --> 00:53:00.120 align:middle line:84% the proteins in this case that are involved 00:53:00.120 --> 00:53:05.380 align:middle line:84% in purine biosynthesis from simpler molecules. 00:53:05.380 --> 00:53:07.360 align:middle line:84% But in other organisms, you might 00:53:07.360 --> 00:53:09.790 align:middle line:84% have them scattered all over the genome, 00:53:09.790 --> 00:53:11.590 align:middle line:90% but they might be co-regulated. 00:53:11.590 --> 00:53:13.970 align:middle line:84% Their RNAs might go up and down together. 00:53:13.970 --> 00:53:17.080 align:middle line:84% And so, if you look at enough genomes, 00:53:17.080 --> 00:53:23.170 align:middle line:84% you can reconstruct the likely combination of enzymes. 00:53:23.170 --> 00:53:25.710 align:middle line:90% And here's how it might work. 00:53:25.710 --> 00:53:27.935 align:middle line:84% In any one of these, for example, E. coli, 00:53:27.935 --> 00:53:30.310 align:middle line:84% you might see that they're scattered about-- a pair here, 00:53:30.310 --> 00:53:31.020 align:middle line:90% and a pair there. 00:53:31.020 --> 00:53:32.983 align:middle line:90% Singletons don't help much. 00:53:32.983 --> 00:53:34.650 align:middle line:84% But if you take all the pairs from a lot 00:53:34.650 --> 00:53:40.260 align:middle line:84% of different organisms, you can reconstruct this network 00:53:40.260 --> 00:53:44.260 align:middle line:84% where you say, oh, this gene will 00:53:44.260 --> 00:53:47.890 align:middle line:84% call L, Q, Y, C-- all these are probably 00:53:47.890 --> 00:53:49.330 align:middle line:90% involved in the same process. 00:53:49.330 --> 00:53:51.668 align:middle line:84% If you get a hint for what any one of them does, 00:53:51.668 --> 00:53:53.835 align:middle line:84% say, one of them is involved in purine biosynthesis, 00:53:53.835 --> 00:53:55.420 align:middle line:84% so then you find that they all are. 00:53:55.420 --> 00:53:59.870 align:middle line:84% And if you guess they might be co-regulated very tightly. 00:53:59.870 --> 00:54:01.840 align:middle line:84% So now let's figure out how we actually 00:54:01.840 --> 00:54:05.840 align:middle line:84% measure that they're co-regulated very tightly. 00:54:05.840 --> 00:54:08.510 align:middle line:90% And the way we can do that-- 00:54:08.510 --> 00:54:12.170 align:middle line:84% as we do that, whatever method we use, we want to ask 00:54:12.170 --> 00:54:15.200 align:middle line:84% are we interested in ratios, relative changes, 00:54:15.200 --> 00:54:17.090 align:middle line:84% or are we interested in absolute values? 00:54:17.090 --> 00:54:20.210 align:middle line:84% There are various things that we can 00:54:20.210 --> 00:54:21.800 align:middle line:84% do with absolute amounts that are 00:54:21.800 --> 00:54:24.460 align:middle line:90% very hard to do with ratios. 00:54:24.460 --> 00:54:28.380 align:middle line:84% In particular, if we want to ask, 00:54:28.380 --> 00:54:34.170 align:middle line:84% is a particular protein level high because its translation is 00:54:34.170 --> 00:54:37.680 align:middle line:84% efficient or is it high because its transcription is efficient, 00:54:37.680 --> 00:54:41.910 align:middle line:84% if you find that it's full of abundant codons 00:54:41.910 --> 00:54:46.170 align:middle line:84% as if it wants to be efficiently translated, is it also 00:54:46.170 --> 00:54:48.570 align:middle line:84% have a high-level promoter as if it wants 00:54:48.570 --> 00:54:51.240 align:middle line:90% to be transcriptionally active? 00:54:51.240 --> 00:54:54.630 align:middle line:84% These sorts of questions really benefit 00:54:54.630 --> 00:54:58.470 align:middle line:84% from having absolute amounts, meaning 00:54:58.470 --> 00:55:00.400 align:middle line:84% so many molecules of RNA per cell, 00:55:00.400 --> 00:55:02.550 align:middle line:84% so many molecules of protein per cell. 00:55:02.550 --> 00:55:04.920 align:middle line:84% But we get that to direct causality, 00:55:04.920 --> 00:55:06.990 align:middle line:90% we want to get at the motifs. 00:55:06.990 --> 00:55:09.960 align:middle line:84% This would be one of the objectives of doing the RNA 00:55:09.960 --> 00:55:14.280 align:middle line:84% quantitation, to allow us to cluster RNAs that 00:55:14.280 --> 00:55:17.430 align:middle line:84% are co-expressed, and then to start looking for motifs 00:55:17.430 --> 00:55:19.140 align:middle line:90% and direct causality. 00:55:19.140 --> 00:55:21.540 align:middle line:84% Another thing that we might want to do is to classify. 00:55:21.540 --> 00:55:26.250 align:middle line:84% We can ask whether small molecules or mutations, 00:55:26.250 --> 00:55:30.060 align:middle line:84% such as occur in cancers, cause enough of a signature 00:55:30.060 --> 00:55:36.110 align:middle line:84% that you can then use as to say, OK, this self state that we see 00:55:36.110 --> 00:55:40.010 align:middle line:84% is a recognizable small molecule effect, or stress effect, 00:55:40.010 --> 00:55:41.585 align:middle line:90% or mutational effect, cancer. 00:55:41.585 --> 00:55:44.280 align:middle line:90% 00:55:44.280 --> 00:55:51.420 align:middle line:84% Now, when we-- we will be talking 00:55:51.420 --> 00:55:55.170 align:middle line:84% about microarray and related methods, 00:55:55.170 --> 00:56:00.390 align:middle line:84% but I want you to question the advantages and disadvantages 00:56:00.390 --> 00:56:01.410 align:middle line:90% of these methods. 00:56:01.410 --> 00:56:05.100 align:middle line:84% And so I'll compare it to a number-- 00:56:05.100 --> 00:56:08.460 align:middle line:84% but let's start with the most dramatic comparison, which 00:56:08.460 --> 00:56:11.370 align:middle line:90% is with in situ hybridization. 00:56:11.370 --> 00:56:15.930 align:middle line:84% So in array hybridization you'll have tens of thousands 00:56:15.930 --> 00:56:21.325 align:middle line:84% of different gene probes mobilized on a solid surface. 00:56:21.325 --> 00:56:23.900 align:middle line:90% 00:56:23.900 --> 00:56:29.530 align:middle line:84% And you'll label up the RNA from a mixture of different cells-- 00:56:29.530 --> 00:56:33.070 align:middle line:84% different mixture of different RNAs within a cell. 00:56:33.070 --> 00:56:34.750 align:middle line:84% But you'll be able to ask questions 00:56:34.750 --> 00:56:36.670 align:middle line:90% about 10,000 genes at a time. 00:56:36.670 --> 00:56:39.070 align:middle line:84% In an in situ experiment, it's the other way around. 00:56:39.070 --> 00:56:43.690 align:middle line:84% You take a cell in it's fairly natural environment, 00:56:43.690 --> 00:56:46.930 align:middle line:84% usually fixed, but fixed with maintaining 00:56:46.930 --> 00:56:49.990 align:middle line:90% the spatial aspects. 00:56:49.990 --> 00:56:53.430 align:middle line:84% Then, if you look within the cell, 00:56:53.430 --> 00:56:56.460 align:middle line:84% with a single gene at a time, or maybe two 00:56:56.460 --> 00:56:59.560 align:middle line:84% or three at a time, a very small number, not tens of thousands, 00:56:59.560 --> 00:57:03.177 align:middle line:84% you can look to see whether the RNA is uniformly 00:57:03.177 --> 00:57:05.760 align:middle line:84% spread throughout the cell and uniformly spread throughout all 00:57:05.760 --> 00:57:08.130 align:middle line:84% the cells in the tissue, or in, say, you've 00:57:08.130 --> 00:57:12.010 align:middle line:84% got a mixed population of yeast cells, whatever. 00:57:12.010 --> 00:57:14.940 align:middle line:84% And you can find cases in the literature 00:57:14.940 --> 00:57:17.970 align:middle line:84% where it is not uniformly present in all the cells 00:57:17.970 --> 00:57:19.680 align:middle line:84% and not even uniformly within a cell. 00:57:19.680 --> 00:57:23.280 align:middle line:84% Here is one of the more dramatic cases where 00:57:23.280 --> 00:57:28.710 align:middle line:84% the two X chromosomes in mammals behave differently 00:57:28.710 --> 00:57:30.180 align:middle line:90% from one another. 00:57:30.180 --> 00:57:35.070 align:middle line:84% The female mammals will have one RNA-- 00:57:35.070 --> 00:57:38.640 align:middle line:84% one chromosome expressing most of its RNAs at normal levels, 00:57:38.640 --> 00:57:41.550 align:middle line:84% and the other chromosome expressing almost no RNAs. 00:57:41.550 --> 00:57:44.670 align:middle line:84% It is expressing at least one RNA, and that RNA is-- 00:57:44.670 --> 00:57:52.860 align:middle line:84% which is XIST and it's covering that whole chromosome 00:57:52.860 --> 00:57:54.720 align:middle line:84% or is localized over that chromosome and not 00:57:54.720 --> 00:57:55.720 align:middle line:90% the rest of the cell. 00:57:55.720 --> 00:57:58.890 align:middle line:84% So this is an extreme case of localization 00:57:58.890 --> 00:58:01.890 align:middle line:84% that you can monitor with microscopic methods, 00:58:01.890 --> 00:58:04.170 align:middle line:90% fluorescent microscopic methods. 00:58:04.170 --> 00:58:07.830 align:middle line:84% Instead, we'll-- keep this in the back of your mind as you 00:58:07.830 --> 00:58:12.780 align:middle line:84% look through the microarray and other experiments where 00:58:12.780 --> 00:58:16.590 align:middle line:84% you're mushing together a variety of cells that might be 00:58:16.590 --> 00:58:18.540 align:middle line:84% in different stages of the cell cycle, 00:58:18.540 --> 00:58:20.530 align:middle line:84% might have slightly different environments, 00:58:20.530 --> 00:58:22.830 align:middle line:84% and even within the cell, the RNA-- 00:58:22.830 --> 00:58:25.680 align:middle line:84% you're losing the information about the RNA localization. 00:58:25.680 --> 00:58:28.270 align:middle line:90% 00:58:28.270 --> 00:58:37.030 align:middle line:84% Let's take a short break and then come back and connect on-- 00:58:37.030 --> 00:58:41.530 align:middle line:84% finish up in situ hybridization, and connect to other methods 00:58:41.530 --> 00:58:44.160 align:middle line:90% for quantitation of RNA. 00:58:44.160 --> 00:58:45.000 align:middle line:90%