WEBVTT 00:00:00.640 --> 00:00:02.459 You know, I was actually thinking about the concept 00:00:02.459 --> 00:00:04.599 of a library the other day. Oh yeah, just the 00:00:04.599 --> 00:00:06.580 general idea of one. Yeah, specifically that 00:00:06.580 --> 00:00:08.699 really old -school image. You know, you picture 00:00:08.699 --> 00:00:11.460 the smell of dust, the silence, rows and rows 00:00:11.460 --> 00:00:14.560 of physical books that supposedly contain everything 00:00:14.560 --> 00:00:17.739 humanity has ever thought or written down. It's 00:00:17.739 --> 00:00:20.820 a really powerful image. It is. It's basically 00:00:20.820 --> 00:00:23.059 the external hard drive of human civilization. 00:00:23.480 --> 00:00:27.129 Exactly. But today, for our deep dive... We're 00:00:27.129 --> 00:00:29.410 going to talk about a library that makes the 00:00:29.410 --> 00:00:31.609 Library of Congress look like, I don't know, 00:00:31.690 --> 00:00:35.030 a distinct stack of Post -it notes. And the crazy 00:00:35.030 --> 00:00:37.530 thing is this library doesn't store books. It 00:00:37.530 --> 00:00:40.670 doesn't store maps. It stores the literal instructions 00:00:40.670 --> 00:00:44.549 for building well, everything alive. Or at least 00:00:44.549 --> 00:00:46.609 everything alive that we've managed to actually 00:00:46.609 --> 00:00:49.549 run through a sequencer so far. Right. We are 00:00:49.549 --> 00:00:51.649 diving into GenBank, and if you're joining us 00:00:51.649 --> 00:00:53.649 for this deep dive, you probably already know 00:00:53.649 --> 00:00:55.950 we love a mind -boggling number to set the stage. 00:00:56.149 --> 00:00:58.250 So to give you a sense of the sheer scale of 00:00:58.250 --> 00:01:01.090 this place right off the dot, as of October 2024, 00:01:01.549 --> 00:01:04.689 this database contained 34 trillion base pairs 00:01:04.689 --> 00:01:07.689 of data. Yeah, 34 trillion. It is a number that 00:01:07.689 --> 00:01:10.549 is genuinely hard to even wrap your head around. 00:01:10.689 --> 00:01:13.099 It's totally abstract. Yeah. Right. I mean, I 00:01:13.099 --> 00:01:15.879 can't picture 34 trillion of anything. But by 00:01:15.879 --> 00:01:17.579 the end of this conversation, we're going to 00:01:17.579 --> 00:01:20.459 make that number feel very real. Because GenBank 00:01:20.459 --> 00:01:23.299 isn't just some cold storage facility for data. 00:01:23.400 --> 00:01:26.540 It is effectively the operating system for modern 00:01:26.540 --> 00:01:29.060 biology. Oh, absolutely. Like if you've eaten 00:01:29.060 --> 00:01:31.700 bread today, taken an antibiotic or had a COVID 00:01:31.700 --> 00:01:34.439 test, your life is intersected with this database. 00:01:34.780 --> 00:01:37.379 That is no exaggeration either. GenBank is the 00:01:37.379 --> 00:01:39.739 gold standard. It's the open access collection 00:01:39.739 --> 00:01:42.280 of all publicly available nucleotide sequences 00:01:42.280 --> 00:01:45.340 and their protein translations. If you're a researcher, 00:01:45.500 --> 00:01:47.620 you aren't just visiting this library. You are 00:01:47.620 --> 00:01:49.680 practically living in it. It's the water the 00:01:49.680 --> 00:01:52.340 scientific community swims in. But here is the 00:01:52.340 --> 00:01:54.280 hook. And really the reason I wanted to pull 00:01:54.280 --> 00:01:57.159 these specific sources for us today. tend to 00:01:57.159 --> 00:01:59.959 think of scientific databases as these pristine, 00:02:00.040 --> 00:02:03.680 infallible vaults of absolute truth. Like, you 00:02:03.680 --> 00:02:05.239 know, if it's in the computer, it must be right. 00:02:05.400 --> 00:02:07.480 And that is a very dangerous assumption to make. 00:02:07.599 --> 00:02:09.900 Because as we dug into the reports on GenBank. 00:02:10.159 --> 00:02:12.819 We found that this library is messy. It's growing 00:02:12.819 --> 00:02:15.400 so incredibly fast, doubling in size every 18 00:02:15.400 --> 00:02:17.819 months, actually, that it's starting to break 00:02:17.819 --> 00:02:20.580 the systems meant to organize it. We've got fake 00:02:20.580 --> 00:02:23.740 news in the form of misidentified fish. We've 00:02:23.740 --> 00:02:25.759 got anonymous turtles that shouldn't exist. And 00:02:25.759 --> 00:02:28.080 we have a quote unquote leaderboard of life that 00:02:28.080 --> 00:02:30.539 puts common bread wheat way above human beings. 00:02:30.659 --> 00:02:33.319 It is a perfect storm of exponential growth and 00:02:33.319 --> 00:02:36.180 human error. It's a testament to human curiosity, 00:02:36.360 --> 00:02:40.020 sure, but also to our fallibility. So today's 00:02:40.020 --> 00:02:43.500 mission is to basically unpack this beast. We're 00:02:43.500 --> 00:02:44.699 going to look at the scale. We're going to look 00:02:44.699 --> 00:02:46.340 at those charts to see who the main characters 00:02:46.340 --> 00:02:48.900 of biological research really are. And we're 00:02:48.900 --> 00:02:51.340 going to talk about why, despite being the scientific 00:02:51.340 --> 00:02:54.120 Bible for so many, it might be harboring some 00:02:54.120 --> 00:02:56.300 pretty significant glitches. It's a tree deep 00:02:56.300 --> 00:02:59.379 dive into the very infrastructure of modern biology. 00:02:59.680 --> 00:03:01.800 So let's start with that number again, 34 trillion 00:03:01.800 --> 00:03:04.379 base pairs. Help us break down the actual inventory 00:03:04.379 --> 00:03:06.750 here. It's difficult because the scale is just 00:03:06.750 --> 00:03:09.229 so vast. But let's look at the distinct sequences. 00:03:09.590 --> 00:03:12.449 As of release 250, which came out in October 00:03:12.449 --> 00:03:17.310 2024, we are looking at over 4 .7 billion distinct 00:03:17.310 --> 00:03:20.090 nucleotide sequences. And just to ensure we're 00:03:20.090 --> 00:03:22.430 perfectly aligned here, when we say nucleotide 00:03:22.430 --> 00:03:24.349 sequences, we're talking about the A, C, T, and 00:03:24.349 --> 00:03:27.229 Gs, right? The raw code of DNA. Correct. The 00:03:27.229 --> 00:03:29.930 base genetic code. And that data covers more 00:03:29.930 --> 00:03:34.800 than 580 ,000 formally described species. So 00:03:34.800 --> 00:03:36.800 from the absolutely smallest bacteria to the 00:03:36.800 --> 00:03:38.580 blue whale, if a scientist has sequenced it, 00:03:38.659 --> 00:03:40.979 it is likely sitting on a server in Maryland 00:03:40.979 --> 00:03:44.680 right now. 580 ,000 species. That is an absurd 00:03:44.680 --> 00:03:47.099 amount of life. But what really jumped out at 00:03:47.099 --> 00:03:49.270 me in the source material was the speed. In the 00:03:49.270 --> 00:03:50.810 tech world, you know, we always talk about Moore's 00:03:50.810 --> 00:03:53.289 Law. Right, the idea that computing power doubles 00:03:53.289 --> 00:03:55.770 roughly every two years, the benchmark for rapid 00:03:55.770 --> 00:03:58.689 progress. Exactly. But GenBank seems to be leaving 00:03:58.689 --> 00:04:01.169 Moore's Law completely in the dust. Yeah, biology 00:04:01.169 --> 00:04:04.129 has its own version of that law, and it is remarkably 00:04:04.129 --> 00:04:06.849 aggressive. Since GenBank started back in 1982, 00:04:07.270 --> 00:04:09.750 the number of bases in the database has doubled 00:04:09.750 --> 00:04:12.229 approximately every 18 months. So it's actually 00:04:12.229 --> 00:04:14.349 growing faster than the computers we use to analyze 00:04:14.349 --> 00:04:17.649 it. In some ways, yes, it is exponential growth 00:04:17.649 --> 00:04:20.689 on steroids. Imagine a physical library where 00:04:20.689 --> 00:04:22.750 every year and a half you literally have to build 00:04:22.750 --> 00:04:25.209 a new wing twice the size of the previous one 00:04:25.209 --> 00:04:28.509 and you fill it instantly. And who is building 00:04:28.509 --> 00:04:30.250 these wings? Like who is actually running this 00:04:30.250 --> 00:04:33.310 thing? Because managing 34 trillion base pairs 00:04:33.310 --> 00:04:35.870 sounds like an administrative nightmare. It is 00:04:35.870 --> 00:04:38.350 a massive coordination effort, as you'd expect. 00:04:38.550 --> 00:04:41.730 It's produced and maintained by the NCBI, which 00:04:41.730 --> 00:04:43.829 is the National Center for Biotechnology Information. 00:04:44.230 --> 00:04:46.689 Which falls under the NIH, the National Institutes 00:04:46.689 --> 00:04:48.790 of Health in the U .S. Correct. But, and this 00:04:48.790 --> 00:04:51.529 is a really crucial distinction, it is not just 00:04:51.529 --> 00:04:54.250 an American project. Science doesn't really respect 00:04:54.250 --> 00:04:57.189 borders and neither does DNA. GenBank is part 00:04:57.189 --> 00:04:59.769 of the International Nucleotide Sequence Database 00:04:59.769 --> 00:05:03.009 Collaboration, or INSDC. That is quite a mouthful. 00:05:03.089 --> 00:05:05.370 It is. But basically it means they are constantly 00:05:05.370 --> 00:05:08.350 synced up with the DNA data bank of Japan and 00:05:08.350 --> 00:05:11.689 the European Nucleotide Archive. So while the 00:05:11.689 --> 00:05:14.310 main servers might be in Maryland, the effort 00:05:14.310 --> 00:05:17.250 is truly global. It's practically a planetary 00:05:17.250 --> 00:05:20.750 brain. Okay, so if the server is the brain, I 00:05:20.750 --> 00:05:22.449 really want to know what it's thinking about 00:05:22.449 --> 00:05:25.459 the most. One of the most fun things we found 00:05:25.459 --> 00:05:27.899 in the source material was this concept of the 00:05:27.899 --> 00:05:31.500 top 20 organisms. The leaderboard of life. Yes, 00:05:31.540 --> 00:05:34.100 exactly. I want to see who the main characters 00:05:34.100 --> 00:05:38.060 are. If we look at the organism with the absolute 00:05:38.060 --> 00:05:42.259 most base pairs stored in GenBank, who takes 00:05:42.259 --> 00:05:44.399 the gold medal? Now, I'm going to go out on a 00:05:44.399 --> 00:05:46.959 limb here. I feel like humans are pretty obsessed 00:05:46.959 --> 00:05:49.740 with ourselves. We sequence the human genome. 00:05:49.860 --> 00:05:52.759 We're constantly studying our own diseases. Surely 00:05:52.759 --> 00:05:55.040 we are number one on this list. That is the most 00:05:55.040 --> 00:05:57.040 reasonable guess you could possibly make. And 00:05:57.040 --> 00:05:59.120 you would be completely wrong. Of course I am. 00:05:59.180 --> 00:06:00.920 We didn't even win our own popularity contest. 00:06:01.180 --> 00:06:03.399 We didn't even make the podium. Homo sapiens 00:06:03.399 --> 00:06:05.339 humans actually come in at number five. We have 00:06:05.339 --> 00:06:08.839 about 27 .8 billion base pairs stored in GenBank. 00:06:09.120 --> 00:06:11.500 Number five. Wow, that is actually kind of humbling. 00:06:11.560 --> 00:06:14.980 Okay, so who beat us? Who is the undisputed heavyweight 00:06:14.980 --> 00:06:17.740 champion of GenBank? That honor goes to Tritacum 00:06:17.740 --> 00:06:20.720 Estabum. Tritacum. That sounds like... Common 00:06:20.720 --> 00:06:22.899 bread wheat. Wheat, this stuff in my toast is 00:06:22.899 --> 00:06:26.019 number one. By an absolute landslide, wheat has 00:06:26.019 --> 00:06:29.379 a massive presence in the database, about 215 00:06:29.379 --> 00:06:32.779 billion base pairs. That is almost 10 times the 00:06:32.779 --> 00:06:35.319 amount of data we have for human beings. Why 00:06:35.319 --> 00:06:38.300 wheat? Is wheat secretly way more complex than 00:06:38.300 --> 00:06:40.490 us? Or are we just that obsessed with carbs? 00:06:40.689 --> 00:06:42.930 Well, it's a bit of both, actually. Genetically, 00:06:43.089 --> 00:06:46.689 wheat is a beast. It has a huge, highly complex 00:06:46.689 --> 00:06:49.069 genome. It's hexaploid, meaning it has six sets 00:06:49.069 --> 00:06:52.129 of chromosomes. But really, this reflects human 00:06:52.129 --> 00:06:54.589 civilization's absolute dependence on agriculture. 00:06:55.050 --> 00:06:57.750 Right. If the global wheat crop fails, we have 00:06:57.750 --> 00:07:00.329 a global catastrophe. Exactly. We are studying 00:07:00.329 --> 00:07:02.810 it constantly for food security, for yield optimization, 00:07:03.110 --> 00:07:05.629 for disease resistance. We are obsessed with 00:07:05.629 --> 00:07:07.529 wheat because our survival fundamentally depends 00:07:07.529 --> 00:07:10.500 on it. scientifically, it gets the lion's share 00:07:10.500 --> 00:07:12.079 of the attention. Okay, that makes total sense. 00:07:12.220 --> 00:07:14.079 Respect the wheat. So who took the silver medal? 00:07:14.160 --> 00:07:15.779 Who's number two? Now, this one tells a very 00:07:15.779 --> 00:07:17.800 specific story about our recent history. It's 00:07:17.800 --> 00:07:20.480 essentially a snapshot of a crisis. Number two 00:07:20.480 --> 00:07:24.870 is SARS -CoV -2. The COVID virus. Exactly. And 00:07:24.870 --> 00:07:27.610 just think about how remarkable that is. This 00:07:27.610 --> 00:07:30.509 is a virus that functionally didn't exist in 00:07:30.509 --> 00:07:32.430 the public consciousness or the database just 00:07:32.430 --> 00:07:36.430 a few years ago. Now it has about 165 billion 00:07:36.430 --> 00:07:40.230 base pairs stored. That is incredible. It really 00:07:40.230 --> 00:07:43.129 visualizes the pivot, doesn't it? It shows the 00:07:43.129 --> 00:07:46.029 exact moment the entire global scientific community 00:07:46.029 --> 00:07:48.930 just dropped what it was doing and stared at 00:07:48.930 --> 00:07:51.529 this one single thing. It really does. It shows 00:07:51.529 --> 00:07:54.209 how GenBank acts as a mirror for scientific priority. 00:07:54.449 --> 00:07:56.730 It doesn't just store biology. It stores our 00:07:56.730 --> 00:07:59.430 reaction to biological crises. When the pandemic 00:07:59.430 --> 00:08:02.269 hit, every lab with a sequencer anywhere in the 00:08:02.269 --> 00:08:04.670 world started uploading viral genomes. So we 00:08:04.670 --> 00:08:06.910 have wheat at number one, COVID at number two. 00:08:06.990 --> 00:08:10.509 Who rounds out the top three? Barley. More grains. 00:08:11.949 --> 00:08:13.709 We really are just farmers with fancy computers, 00:08:13.870 --> 00:08:15.850 aren't we? Humans like their beer and their bread. 00:08:15.889 --> 00:08:17.689 It's pretty undeniable at this point. Okay, so 00:08:17.689 --> 00:08:20.470 wheat, the virus, barley. Then the lab mouse 00:08:20.470 --> 00:08:23.290 at number four, I assume. Yes, most musculus 00:08:23.290 --> 00:08:26.379 is number four. sitting right there just ahead 00:08:26.379 --> 00:08:29.279 of humans. And if you go further down the list, 00:08:29.339 --> 00:08:31.620 past humans at number five, you see the other 00:08:31.620 --> 00:08:34.720 standard research staples. You have E. coli at 00:08:34.720 --> 00:08:37.240 number seven. The absolute workhorse of the lab. 00:08:37.399 --> 00:08:40.200 Exactly. You have the zebrafish, Danio Rario, 00:08:40.340 --> 00:08:43.600 at number nine. Dogs and pigs also make the top 00:08:43.600 --> 00:08:45.860 20 list. It's really interesting that the list 00:08:45.860 --> 00:08:48.679 isn't necessarily the most complex animals or 00:08:48.679 --> 00:08:51.080 the coolest animals. Like, I don't see any lions 00:08:51.080 --> 00:08:53.279 or tigers on there. It's literally just the animals 00:08:53.279 --> 00:08:55.840 we need. Precisely. It is a pure reflection of 00:08:55.840 --> 00:08:58.059 human utility. We sequence what we eat, what 00:08:58.059 --> 00:08:59.860 makes us sick, and what lives in our houses. 00:09:00.120 --> 00:09:02.279 It's a very pragmatic list when you look at it. 00:09:02.399 --> 00:09:04.919 So how did this all begin? You mentioned 1982 00:09:04.919 --> 00:09:08.100 earlier. That feels like the Stone Age for computers. 00:09:08.240 --> 00:09:10.899 I'm picturing giant tape drives in those green 00:09:10.899 --> 00:09:13.820 tech screens. You're not far off at all. And 00:09:13.820 --> 00:09:15.580 the origin story is actually quite surprising 00:09:15.580 --> 00:09:17.399 because it starts right in the shadow of the 00:09:17.399 --> 00:09:19.820 Cold War. Really? Yeah. How do you get from the 00:09:19.820 --> 00:09:23.240 Cold War to DNA sequencing? It traces back to 00:09:23.240 --> 00:09:25.460 the Los Alamos National Laboratory. Wait, the 00:09:25.460 --> 00:09:28.149 atomic bomb place. The very same, specifically 00:09:28.149 --> 00:09:31.250 the theoretical biology and biophysics group 00:09:31.250 --> 00:09:34.029 there. There was a physicist named Walter Goad, 00:09:34.090 --> 00:09:37.250 who is essentially the father of GenBank. I suppose 00:09:37.250 --> 00:09:39.509 if you have the massive computing power required 00:09:39.509 --> 00:09:41.610 to calculate nuclear blast radiuses, you might 00:09:41.610 --> 00:09:43.789 as well use it for biology, too. That is exactly 00:09:43.789 --> 00:09:46.870 the idea. The math of pattern recognition overlaps 00:09:46.870 --> 00:09:49.960 more than you'd think. So in 1982, with funding 00:09:49.960 --> 00:09:52.379 from the NIH, the National Science Foundation, 00:09:52.700 --> 00:09:54.960 the DOE, and even the Department of Defense, 00:09:55.240 --> 00:09:57.940 they officially launched GenBank. They must have 00:09:57.940 --> 00:10:00.440 been tiny back then. Minuscule. By the end of 00:10:00.440 --> 00:10:03.539 1983, they had stored just over 2 ,000 sequences. 00:10:03.940 --> 00:10:07.669 2 ,000. And now we're at 4 .7 billion. That is 00:10:07.669 --> 00:10:10.250 quite the growth spurt. It was a very different 00:10:10.250 --> 00:10:12.470 world. In the mid -80s, the project was actually 00:10:12.470 --> 00:10:15.049 managed by a bioinformatics company at Stanford 00:10:15.049 --> 00:10:17.750 called Intelligenetics. It didn't actually move 00:10:17.750 --> 00:10:20.350 to the NCBI until that transition period between 00:10:20.350 --> 00:10:23.909 1989 and 1992. And I read in the sources that 00:10:23.909 --> 00:10:26.169 this project actually helped kickstart the way 00:10:26.169 --> 00:10:28.570 scientists talk to each other online. Is that 00:10:28.570 --> 00:10:32.230 right? Yes, the Bio -CI or BioNet newsgroups. 00:10:32.230 --> 00:10:35.230 It was one of the earliest examples of open access 00:10:35.230 --> 00:10:38.429 communication among scientists. Before the modern 00:10:38.429 --> 00:10:41.190 internet as we know it even existed, GenBank 00:10:41.190 --> 00:10:43.909 was already fostering this culture of share what 00:10:43.909 --> 00:10:47.110 you find. It created the template for open science. 00:10:47.470 --> 00:10:50.039 Which brings us to the mechanics of it. How does 00:10:50.039 --> 00:10:52.720 the data actually get into the library? Because 00:10:52.720 --> 00:10:54.580 I think there's a common misconception that there 00:10:54.580 --> 00:10:57.559 is some librarian of life sitting at a desk at 00:10:57.559 --> 00:11:00.480 the NIH scanning books and carefully typing in 00:11:00.480 --> 00:11:02.539 codes. Yeah, that would be physically impossible 00:11:02.539 --> 00:11:05.779 given the volume. GenBank relies entirely on 00:11:05.779 --> 00:11:08.259 direct submissions. So it's user -generated content. 00:11:08.860 --> 00:11:12.100 Like the Wikipedia of DNA? In a sense, yes. If 00:11:12.100 --> 00:11:14.779 you are a researcher in a lab in Brazil or a 00:11:14.779 --> 00:11:17.460 grad student in Tokyo, you use web -based tools 00:11:17.460 --> 00:11:20.679 like BankIt or a program called Table2Azen. You 00:11:20.679 --> 00:11:22.419 essentially fill out a form, attach your sequence 00:11:22.419 --> 00:11:24.259 data, and send it off. And what about the big 00:11:24.259 --> 00:11:26.480 sequencing centers, the ones churning out terabytes 00:11:26.480 --> 00:11:28.700 of data a day? They have automated pipelines 00:11:28.700 --> 00:11:31.759 doing bulk submissions 24 -7. It's a constant, 00:11:31.840 --> 00:11:35.159 unending stream of data. So say I upload my sequence. 00:11:36.059 --> 00:11:38.700 Does it go straight to the public? Or is there 00:11:38.700 --> 00:11:41.299 a bouncer at the door checking IDs? There is 00:11:41.299 --> 00:11:44.139 a vetting process, but it's specific. GenBank 00:11:44.139 --> 00:11:46.580 staff does examine the originality of the data. 00:11:46.759 --> 00:11:49.559 They perform quality assurance checks to make 00:11:49.559 --> 00:11:51.820 sure the file isn't corrupted and that it makes 00:11:51.820 --> 00:11:54.720 basic biological sense. Once it passes that... 00:11:54.940 --> 00:11:56.940 They assign it an accession number. Like a Dewey 00:11:56.940 --> 00:11:58.740 decimal number. Think of it more like a social 00:11:58.740 --> 00:12:01.500 security number for that specific piece of data. 00:12:01.799 --> 00:12:04.200 It's a unique identifier so other scientists 00:12:04.200 --> 00:12:06.759 can find it and cite it in their papers. Then 00:12:06.759 --> 00:12:09.200 it's released to the public database. You can 00:12:09.200 --> 00:12:12.559 download it via FTP or search for it using a 00:12:12.559 --> 00:12:15.000 tool called Entrez. And the cost for all this? 00:12:15.240 --> 00:12:18.139 Free. It is totally open access. Anyone with 00:12:18.139 --> 00:12:19.980 an internet connection can look at the code of 00:12:19.980 --> 00:12:23.139 life. However. And this is a nuance people often 00:12:23.139 --> 00:12:25.299 miss when we talk about this. I saw in the notes 00:12:25.299 --> 00:12:27.879 that open access doesn't necessarily mean free 00:12:27.879 --> 00:12:30.659 of strings. That is absolutely correct. The NCBI 00:12:30.659 --> 00:12:33.139 places no restrictions on the use of the data, 00:12:33.259 --> 00:12:35.120 but that doesn't mean the data is completely 00:12:35.120 --> 00:12:38.100 free of intellectual property claims. What do 00:12:38.100 --> 00:12:40.139 you mean by that? Well, some submitters might 00:12:40.139 --> 00:12:42.980 claim patents or copyrights on the specific data 00:12:42.980 --> 00:12:46.159 they submit. NCBI explicitly says they are not 00:12:46.159 --> 00:12:48.740 in a position to police those claims. So just 00:12:48.740 --> 00:12:50.600 because you found it in the public library doesn't 00:12:50.600 --> 00:12:52.460 mean you can necessarily use it to make a commercial 00:12:52.460 --> 00:12:54.779 product without checking the fine print first. 00:12:55.080 --> 00:12:57.399 That is a really fascinating legal gray area. 00:12:58.000 --> 00:13:00.980 Okay, so we have this massive, open, rapidly 00:13:00.980 --> 00:13:03.139 growing library. But this is where I want to 00:13:03.139 --> 00:13:05.500 get a little critical. This is the deep dive 00:13:05.500 --> 00:13:07.500 after all. Right, you're anticipating the garbage 00:13:07.500 --> 00:13:09.919 in problem. Exactly. If this library is built 00:13:09.919 --> 00:13:13.139 almost entirely on user submissions and users 00:13:13.139 --> 00:13:16.419 are human, well, humans make mistakes. Can we 00:13:16.419 --> 00:13:18.620 actually trust everything in GenBank? The short 00:13:18.620 --> 00:13:20.840 answer is mostly, but definitely not blindly. 00:13:21.550 --> 00:13:23.970 Because GenBank relies on the scientific community 00:13:23.970 --> 00:13:26.870 to submit data, it naturally also inherits the 00:13:26.870 --> 00:13:28.809 scientific community's errors. Let's look at 00:13:28.809 --> 00:13:30.350 some examples from the sources, because some 00:13:30.350 --> 00:13:32.549 of these are wild. I saw a story about a fish 00:13:32.549 --> 00:13:34.970 that was having a serious identity crisis. Ah, 00:13:35.110 --> 00:13:38.629 yes. The infamous case of Nemipterus mesoprean. 00:13:38.769 --> 00:13:41.850 Which is what kind of fish? It's a type of threadfin 00:13:41.850 --> 00:13:44.830 bream, a pretty commercially important fish in 00:13:44.830 --> 00:13:47.269 the Indo -Pacific region. Okay, a threadfin bream. 00:13:47.330 --> 00:13:50.149 What exactly happened to it? A recent study looked 00:13:50.149 --> 00:13:52.529 at the mitochondrial sequences for this fish 00:13:52.529 --> 00:13:55.809 in GenBank, specifically the cytochrome -sick 00:13:55.809 --> 00:13:58.990 oxidase subunit I sequences used to identify 00:13:58.990 --> 00:14:02.429 the species. They found that 75 % of the sequences 00:14:02.429 --> 00:14:06.710 assigned to this species were wrong. 75%. That's 00:14:06.710 --> 00:14:09.529 not a margin of error. That's just wrong. It's 00:14:09.529 --> 00:14:12.590 a comprehensively failed test. It is a massive 00:14:12.590 --> 00:14:14.730 error rate. And the reason why it happened is 00:14:14.730 --> 00:14:17.629 the real insight here. It highlights how the 00:14:17.629 --> 00:14:20.220 system... creates a feedback loop. Walk us through 00:14:20.220 --> 00:14:22.480 that. How does a fish get that wrong in a processional 00:14:22.480 --> 00:14:25.259 database? Imagine Researcher A catches a fish. 00:14:25.440 --> 00:14:28.360 They misidentify it as a threadfin bream, sequence 00:14:28.360 --> 00:14:30.779 it, and upload that sequence to GenBank. Now, 00:14:30.820 --> 00:14:32.960 Researcher B catches a similar -looking fish. 00:14:33.080 --> 00:14:35.059 They aren't totally sure what it is, so they 00:14:35.059 --> 00:14:37.620 blast the DNA against GenBank to check. And they 00:14:37.620 --> 00:14:39.799 get a match from Researcher A's upload. Exactly. 00:14:39.860 --> 00:14:42.539 They see the match and say, aha, this must be 00:14:42.539 --> 00:14:44.820 Mipterus misoprion because the database says 00:14:44.820 --> 00:14:47.379 so. Then they upload their data confirming it. 00:14:47.519 --> 00:14:49.399 So they are matching their catch against a lie. 00:14:49.840 --> 00:14:53.289 Precisely. It creates a chain of errors. Citation 00:14:53.289 --> 00:14:55.610 laundering. Essentially, that becomes very hard 00:14:55.610 --> 00:14:57.490 to break because everyone is citing the same 00:14:57.490 --> 00:15:00.309 bad data. You end up with a mountain of evidence 00:15:00.309 --> 00:15:02.789 that is essentially built on sand. That is wild. 00:15:03.090 --> 00:15:05.309 It's like copying off the smart kid's homework, 00:15:05.549 --> 00:15:08.309 but the smart kid actually failed the test, and 00:15:08.309 --> 00:15:10.690 now the entire class is failing. And it's not 00:15:10.690 --> 00:15:12.889 just fish. There was a manuscript looking at 00:15:12.889 --> 00:15:16.210 birds, specifically cytochrome B records, that 00:15:16.210 --> 00:15:19.470 showed 45 % of the erroneous records lacked a 00:15:19.470 --> 00:15:22.169 voucher specimen. A voucher specimen. That sounds... 00:15:22.190 --> 00:15:24.830 like a coupon? In biology, a voucher specimen 00:15:24.830 --> 00:15:27.769 is the physical backup. It's the actual preserved 00:15:27.769 --> 00:15:31.490 bird or fish or plant sitting in a museum drawer 00:15:31.490 --> 00:15:33.909 somewhere. It's the physical receipt. So if the 00:15:33.909 --> 00:15:36.210 data looks weird online, you can physically go 00:15:36.210 --> 00:15:38.830 to the drawer and look at the actual bird? Ideally, 00:15:38.870 --> 00:15:41.289 yes. But if you don't have the physical bird 00:15:41.289 --> 00:15:44.690 and the digital record is weird, you have absolutely 00:15:44.690 --> 00:15:47.470 no way to prove if it is a brilliant new discovery 00:15:47.470 --> 00:15:49.929 or just a careless mistake. So those records 00:15:49.929 --> 00:15:52.200 are just ghosts in the machine. In a way, yeah. 00:15:52.360 --> 00:15:54.820 They are unverified beta points floating in the 00:15:54.820 --> 00:15:57.539 system forever. And then you have the anonymous 00:15:57.539 --> 00:16:00.019 problem. This is the turtle subject X issue we 00:16:00.019 --> 00:16:03.500 read about. Right. Often, researchers will submit 00:16:03.500 --> 00:16:06.100 a sequence before they have formally named the 00:16:06.100 --> 00:16:09.299 species. They might call it something like Pelomedusa 00:16:09.299 --> 00:16:15.120 SBS, a CK 2014. Catchy name. Rolls right off 00:16:15.120 --> 00:16:17.529 the tongue. Very. But they do this to get the 00:16:17.529 --> 00:16:20.370 data out there quickly, or to support a draft 00:16:20.370 --> 00:16:22.590 paper they're writing. The problem is, three 00:16:22.590 --> 00:16:25.090 years later, they publish the paper, they officially 00:16:25.090 --> 00:16:28.570 name the turtle... Pila Medusa variabilis, but... 00:16:28.570 --> 00:16:30.690 Don't tell me. They often just forget to go back 00:16:30.690 --> 00:16:33.009 to GenBank and update the original record. So 00:16:33.009 --> 00:16:35.570 the library is full of books with temporary titles 00:16:35.570 --> 00:16:37.769 that never get changed to the real ones. Exactly. 00:16:37.769 --> 00:16:40.090 It causes ongoing confusion because you end up 00:16:40.090 --> 00:16:42.129 with duplicate entries under different names. 00:16:42.330 --> 00:16:44.350 A researcher might think they've found a brand 00:16:44.350 --> 00:16:46.370 new species, but they are actually just looking 00:16:46.370 --> 00:16:48.990 at an unupdated clerical error from five years 00:16:48.990 --> 00:16:52.070 ago. Now, for a biologist studying turtle evolution, 00:16:52.330 --> 00:16:55.019 I get that that's annoying. But does this have 00:16:55.019 --> 00:16:57.779 real -world stakes? Like, what if I'm a doctor? 00:16:58.000 --> 00:17:00.159 Let's say I have a patient with a weird infection, 00:17:00.299 --> 00:17:03.740 and I sequence the bacteria and check GenBank 00:17:03.740 --> 00:17:07.339 to identify it. Could this be dangerous? It could 00:17:07.339 --> 00:17:10.740 be if you rely only on GenBank. That is the critical 00:17:10.740 --> 00:17:13.440 warning here. For clinical identification, like 00:17:13.440 --> 00:17:15.960 blood cultures, where a patient's life might 00:17:15.960 --> 00:17:18.859 be on the line, experts highly recommend not 00:17:18.859 --> 00:17:21.380 putting all your eggs in the GenBank basket. 00:17:21.640 --> 00:17:23.670 What should they do instead? You combine it with 00:17:23.670 --> 00:17:26.509 other more heavily curated databases, things 00:17:26.509 --> 00:17:29.329 like Aztaxany or BB. Like boutique libraries. 00:17:29.670 --> 00:17:32.130 Exactly. These are smaller but much stricter. 00:17:32.250 --> 00:17:34.029 Someone has actually checked the books thoroughly. 00:17:34.549 --> 00:17:36.930 GenBank is the giant warehouse. It has everything, 00:17:37.009 --> 00:17:39.069 but it's messy. These others are the carefully 00:17:39.069 --> 00:17:41.170 curated collections. That makes a lot of sense. 00:17:41.369 --> 00:17:43.849 It seems like the tradeoff for having the biggest, 00:17:44.049 --> 00:17:46.650 most comprehensive library in the world is that 00:17:46.650 --> 00:17:48.250 you're just inevitably going to have some graffiti 00:17:48.250 --> 00:17:50.470 in the margins. That is a very fair way to put 00:17:50.470 --> 00:17:53.750 it. You need the warehouse for discovery, for 00:17:53.750 --> 00:17:56.049 the big picture, for the exponential growth. 00:17:56.289 --> 00:17:59.089 But you definitely need the boutique collections 00:17:59.089 --> 00:18:02.390 for precision. So let's wrap this up. We have 00:18:02.390 --> 00:18:07.190 GenBank. It is a $34 trillion base pair behemoth. 00:18:07.690 --> 00:18:10.670 It puts wheat and COVID above human beings in 00:18:10.670 --> 00:18:13.390 terms of raw data volume. It was born in the 00:18:13.390 --> 00:18:16.009 Cold War and now it lives in the cloud, sinking 00:18:16.009 --> 00:18:18.250 globally every day. And it is still doubling 00:18:18.250 --> 00:18:21.049 every 18 months. It truly is the backbone of 00:18:21.049 --> 00:18:23.569 modern biology. But it's also a stark reminder 00:18:23.569 --> 00:18:25.869 that data isn't always the exact same thing as 00:18:25.869 --> 00:18:28.849 truth. That is the key takeaway for me. GenBank 00:18:28.849 --> 00:18:31.569 is an absolutely essential tool for open science. 00:18:31.730 --> 00:18:33.930 We couldn't do modern medicine without it. But 00:18:33.930 --> 00:18:36.349 we have to remember, it is a record of submissions. 00:18:36.630 --> 00:18:38.549 It reflects what scientists think they found 00:18:38.549 --> 00:18:41.170 at that exact moment in time. It is a living, 00:18:41.349 --> 00:18:43.470 breathing, and occasionally mistaken history 00:18:43.470 --> 00:18:45.930 of our understanding of life. Absolutely. And, 00:18:45.950 --> 00:18:47.710 you know, it raises a pretty provocative question 00:18:47.710 --> 00:18:50.509 for anyone listening. Well, we are currently 00:18:50.509 --> 00:18:53.250 building the future of biology. We are training 00:18:53.250 --> 00:18:56.049 complex AI models on this data. We are designing 00:18:56.049 --> 00:18:58.250 synthetic life. We are hunting for new drugs. 00:18:59.180 --> 00:19:01.640 But if this database contains millions of anonymous 00:19:01.640 --> 00:19:04.319 sequences or mislabeled records that are rarely 00:19:04.319 --> 00:19:06.740 updated or purged. Are we building our future 00:19:06.740 --> 00:19:10.299 on a cracked foundation? Exactly. Are we inadvertently 00:19:10.299 --> 00:19:13.279 baking fiction right into our biological reality? 00:19:13.700 --> 00:19:16.460 And as this library grows faster than any human 00:19:16.460 --> 00:19:18.819 can possibly read, doubling every year and a 00:19:18.819 --> 00:19:21.720 half, how do we ever hope to go back and clean 00:19:21.720 --> 00:19:23.980 it up? We might just be creating a web of knowledge 00:19:23.980 --> 00:19:28.019 so complex and interlinked with tiny errors that 00:19:28.019 --> 00:19:30.660 we can. literally never fully untangle it. It's 00:19:30.660 --> 00:19:33.299 a very distinct possibility. Well, on that slightly 00:19:33.299 --> 00:19:35.440 existential note, I'm going to go look at a sandwich 00:19:35.440 --> 00:19:37.599 with a lot more respect from now on. All hail 00:19:37.599 --> 00:19:39.980 the wheat genome. Number one on the leaderboard, 00:19:40.119 --> 00:19:42.819 number one in our hearts. Thanks for joining 00:19:42.819 --> 00:19:45.000 us on this deep dive into the library of life. 00:19:45.160 --> 00:19:46.200 We'll catch you in the next one.