Latest / Razib Khan's Unsupervised Learning / Daniel Tabin: ancient DNA, the good, bad and ugly
Transcript
- 0:00This podcast is brought to you by the Albany Public Library
- 0:03Main Branch and the
- 0:04SPEAKER_01: generosity of listeners like you.
- 0:09SPEAKER_01: What is a podcast?
- 0:16SPEAKER_00: God, Daddy, these people talk as much as you do.
- 0:26SPEAKER_01: Razeeb Khan's Unsupervised Learning
- 0:29SPEAKER_01: Hey, everybody.
- 0:29SPEAKER_01: It's Razeeb Khan here with the Unsupervised Learning Podcast.
- 0:32SPEAKER_01: And I am here today with soon-to-be Dr. Daniel Tabin.
- 0:36SPEAKER_01: Daniel is in David Wright's lab.
- 0:37Can you tell us a little bit more about yourself and your
- 0:39research,
- 0:39Daniel?
- 0:40SPEAKER_00: Yeah, well, we'll see.
- 0:41SPEAKER_00: Hopefully soon to be.
- 0:42SPEAKER_00: And not that soon.
- 0:43SPEAKER_00: But yeah, I'm a fifth-year grad student.
- 0:47SPEAKER_00: And like David, I study ancient DNA.
- 0:49SPEAKER_00: And sort of my thesis will be on the history of Central Asia and Siberia.
- 0:55That's sort of the thing that connects the main things I've
- 0:57been interested in.
- 0:59SPEAKER_00: And I've done other things,
- 1:00SPEAKER_00: including a recent-ish,
- 1:03like last year paper that's sort of a technical comment
- 1:07challenging another paper
- 1:10SPEAKER_00: on ancient China.
- 1:12SPEAKER_01: Okay.
- 1:12Well,
- 1:12so also I will recommend for the listeners out there,
- 1:15check out Daniel on Twitter.
- 1:17SPEAKER_01: I'll put the link in there.
- 1:18SPEAKER_01: Really good science communicator.
- 1:20SPEAKER_01: He has not fled to blue sky.
- 1:23You know,
- 1:24there are just,
- 1:24it's a lot of science Twitter has kind of like disappeared.
- 1:26And,
- 1:26you know,
- 1:27you're one of the people that I still see in the genetic space.
- 1:30SPEAKER_01: You, maybe Sasha Goose, have a few other people.
- 1:33SPEAKER_01: So I wanted first to talk about something technical.
- 1:35Then maybe we'll get to ancient DNA,
- 1:37maybe some of your other research and
- 1:38SPEAKER_01: you know, the results section, so to speak.
- 1:40SPEAKER_01: So the paper I want to talk about is more of a comment section,
- 1:42SPEAKER_01: not a method section paper,
- 1:43SPEAKER_01: right?
- 1:44Concerned about ancient DNA sequences reported from place to
- 1:47see an individual in
- 1:47SPEAKER_01: Southeast Asia.
- 1:49SPEAKER_01: And so that came out last year.
- 1:50SPEAKER_01: And I'm assuming you've been looking at that longer than 2025.
- 1:55SPEAKER_01: And it's about the red deer cave, which I've heard about.
- 1:57SPEAKER_01: That's not like South China, right?
- 1:59SPEAKER_00: Yeah,
- 2:00SPEAKER_00: it's somewhat famous because,
- 2:02SPEAKER_00: and this probably prompted at least some of the research on the sample,
- 2:08SPEAKER_00: it's like the skulls look very interesting phenotypically,
- 2:12and they were even potentially thought to be kind of archaic
- 2:14hybrids before the
- 2:16SPEAKER_00: ancient DNA revolution.
- 2:20SPEAKER_00: And, yeah, I started looking at it actually in 2020.
- 2:24SPEAKER_00: I read the paper, which makes very interesting claims.
- 2:27SPEAKER_00: And I sort of wanted to forward them.
- 2:30SPEAKER_00: I thought this may be related to the publishing Y signal.
- 2:34I should say maybe that the publishing Y signal is the idea
- 2:37that certain Native
- 2:38SPEAKER_00: Americans,
- 2:39SPEAKER_00: specifically those in Brazil,
- 2:40have some sort of affinity to certain Southeast Asian groups
- 2:43like the Ongay,
- 2:44SPEAKER_00: sort of Andaman Islanders,
- 2:45SPEAKER_00: and potentially Poplins.
- 2:47SPEAKER_00: And there's
- 2:49The MZR paper suggests that there was a migration from
- 2:53Southeast Asia into
- 2:55SPEAKER_00: Northeast Asia,
- 2:56SPEAKER_00: eventually reaching the Americas.
- 2:57SPEAKER_00: And I looked at the paper and I basically,
- 3:01SPEAKER_00: I guess to spoil it,
- 3:03SPEAKER_00: I realized that the claims didn't really make any sense.
- 3:05SPEAKER_00: And kind of more concerningly,
- 3:06SPEAKER_00: even than the claims not really being super substantiated,
- 3:11SPEAKER_00: was that the data was like highly problematic.
- 3:14SPEAKER_00: It's sort of a way that sort of,
- 3:15SPEAKER_00: in both ways are sort of typical and ways that are atypical,
- 3:17SPEAKER_00: I guess.
- 3:18SPEAKER_00: Uh, so I, I, I then, um, tried to figure out what happened.
- 3:21SPEAKER_00: Um,
- 3:22SPEAKER_00: I also want to say before I say anything,
- 3:23SPEAKER_00: I just want to,
- 3:24SPEAKER_00: I want to make sure I get this out that the,
- 3:25SPEAKER_00: the authors of that paper,
- 3:27SPEAKER_00: um,
- 3:28SPEAKER_00: are just extremely,
- 3:29SPEAKER_00: uh,
- 3:30SPEAKER_00: kind and actually like very,
- 3:32SPEAKER_00: I would say collaborative in some way.
- 3:33SPEAKER_00: I mean,
- 3:33SPEAKER_00: I,
- 3:34SPEAKER_00: I,
- 3:34SPEAKER_00: they didn't like directly contribute to the notes,
- 3:36SPEAKER_00: but like I sent them the notes and I felt that they were,
- 3:39SPEAKER_00: even though they,
- 3:39SPEAKER_00: they didn't
- 3:41SPEAKER_00: necessarily say, oh, we're 100% wrong at any point.
- 3:45SPEAKER_00: They gave feedback very friendly.
- 3:47I've heard of other stories where people are trying to critique
- 3:49science and people
- 3:50SPEAKER_00: are very aggressive and rude and mean.
- 3:53SPEAKER_00: They were very gracious.
- 3:55SPEAKER_00: I wanted to actually shout them out.
- 3:57SPEAKER_00: Even though I disagree with their conclusions,
- 4:02SPEAKER_00: I think that the data is problematic and I think it shouldn't be used.
- 4:04SPEAKER_00: I actually think the authors are wonderful people.
- 4:06SPEAKER_00: I wanted to make sure I said that.
- 4:08SPEAKER_01: Yeah, I mean, that's how science should be done.
- 4:11SPEAKER_01: But let's be candid.
- 4:13SPEAKER_01: That's often not how it is done.
- 4:15SPEAKER_01: People get very attached to their results, especially for cool results.
- 4:19I'm going to tell a real quick story because this is a long
- 4:24time ago and people
- 4:24SPEAKER_01: aren't going to know who this is.
- 4:26SPEAKER_01: When I was in graduate school, I had a close friend who was in a lab.
- 4:30SPEAKER_01: And it was,
- 4:31SPEAKER_01: I'm gonna just say it's a plant genetics lab,
- 4:33SPEAKER_01: it's a prominent one,
- 4:36SPEAKER_01: and a few people are gonna be able to guess who this is.
- 4:38SPEAKER_01: But the TLDR is,
- 4:40SPEAKER_01: my friend was a first year graduate student in the lab,
- 4:44and he could not replicate a science publication from like two
- 4:48years before I think
- 4:49SPEAKER_01: he joined the lab.
- 4:50SPEAKER_01: And he became convinced that it was fraud.
- 4:54SPEAKER_01: And,
- 4:55SPEAKER_01: you know,
- 4:55SPEAKER_01: they didn't explicitly ever say it was fraud,
- 4:57SPEAKER_01: but the primary investigator,
- 4:59SPEAKER_01: the professor,
- 4:59SPEAKER_01: eventually did withdraw this science publication.
- 5:04SPEAKER_01: It's a science publication.
- 5:06SPEAKER_01: That's tough, right?
- 5:08And so,
- 5:09you know,
- 5:09from the outside,
- 5:10it's like,
- 5:10okay,
- 5:10this is how it should have worked.
- 5:11SPEAKER_01: But I will tell you, my friend...
- 5:17had to drag the professor to that retraction okay uh he
- 5:23threatened to quit the lab
- 5:24uh because he was not going to work on a project where he knew
- 5:29that the previous
- 5:30stuff he would have to cite uh was probably fraudulent you know
- 5:34and so uh you know
- 5:36if when you see how the sausage is made uh it's tough it's a
- 5:39human enterprise and
- 5:41SPEAKER_01: so we definitely uh you know
- 5:43SPEAKER_01: should give a shout out to people who are collaborative in this way.
- 5:46SPEAKER_01: And,
- 5:46SPEAKER_01: you know,
- 5:47SPEAKER_01: you know,
- 5:47SPEAKER_01: Daniel,
- 5:48SPEAKER_01: data submitted upon request is often more theoretical than reality.
- 5:53I've seen people who've actually tried to do this and half the
- 5:57time the data is
- 5:58SPEAKER_01: just not given to them.
- 5:59SPEAKER_00: Yeah, that's actually a great example from this exact situation.
- 6:02SPEAKER_00: Again,
- 6:03SPEAKER_00: while they didn't necessarily accept the claims,
- 6:05SPEAKER_00: they actually kind of wrote a rebuttal to my paper,
- 6:07SPEAKER_00: which people should read.
- 6:08SPEAKER_00: I disagree with it,
- 6:09SPEAKER_00: obviously,
- 6:10but people should read the rebuttal they wrote to my rebuttal
- 6:13in some way.
- 6:14SPEAKER_00: But actually, they gave me the fast queue files.
- 6:16The fast queue files weren't published,
- 6:19just like the BAM,
- 6:20that data's available to anyone.
- 6:22But I actually have the fast queues,
- 6:24the original reads,
- 6:25which is extremely generous of them.
- 6:27SPEAKER_00: And if you're...
- 6:30SPEAKER_00: going to a war with someone.
- 6:31SPEAKER_00: They were essentially giving me more ammo in some way.
- 6:37SPEAKER_01: They were letting you take them down.
- 6:39SPEAKER_01: It doesn't sound like they slow-walked it.
- 6:41SPEAKER_01: They made it easy for you to make your case, even if they disagreed.
- 6:44SPEAKER_01: That's how it should be done.
- 6:45SPEAKER_01: I think that they disagreed, but I think that they...
- 6:47SPEAKER_00: I think they did continue to agree with their conclusions,
- 6:52SPEAKER_00: but they ultimately were also in favor of the process.
- 6:55SPEAKER_00: And I strongly commend them because I've seen it myself,
- 6:59SPEAKER_00: not in personal experiences,
- 7:00SPEAKER_00: but people I know are like,
- 7:02SPEAKER_00: yeah,
- 7:02you're totally right that academics are often very want to
- 7:06admit they're wrong,
- 7:07SPEAKER_00: unfortunately.
- 7:08SPEAKER_01: Hey, they're human.
- 7:09SPEAKER_01: They're human.
- 7:10So going to this paper and this whole issue of contamination,
- 7:13so you said FASTQ files.
- 7:15Some of the listeners of this podcast are actually biologists,
- 7:19so they'll know what that means.
- 7:20SPEAKER_01: So just to be clear to the listener,
- 7:23SPEAKER_01: when you have whole genome sequences,
- 7:25SPEAKER_01: you get these raw reads,
- 7:27SPEAKER_01: right?
- 7:27SPEAKER_01: These FASTQ files.
- 7:29SPEAKER_01: And then you take the fastfuge files,
- 7:30SPEAKER_01: you reassemble them into something called like a BAM or now a CRAM file,
- 7:34SPEAKER_01: I guess,
- 7:35SPEAKER_01: which is,
- 7:35SPEAKER_01: you know,
- 7:36SPEAKER_01: condenses it and kind of cleans it up.
- 7:38SPEAKER_01: And there's like,
- 7:38SPEAKER_01: you know,
- 7:39SPEAKER_01: data,
- 7:39SPEAKER_01: you know,
- 7:40SPEAKER_01: deduplication,
- 7:41SPEAKER_01: all of these things that you do the fastfuge files.
- 7:43And then you take the BAM and you often reduce that down to
- 7:46just the variants.
- 7:47SPEAKER_01: So like, you know, the VCF file, right?
- 7:50SPEAKER_01: Variant color format.
- 7:51SPEAKER_01: And from that,
- 7:53SPEAKER_01: you often,
- 7:54SPEAKER_01: like in your lab,
- 7:55SPEAKER_01: David Reich's lab,
- 7:56SPEAKER_01: often you snip arrays,
- 7:57SPEAKER_01: which are somewhat different.
- 7:58And those are just the single nucleotide variants that are on
- 8:02an array,
- 8:03on a DNA chip.
- 8:04SPEAKER_01: And so when I write about ancient DNA,
- 8:07SPEAKER_01: mostly I'm writing about the results,
- 8:09SPEAKER_01: and I'm interpreting through the historical lens and whatever.
- 8:11SPEAKER_01: But there's a lot of technical detail on how we get that data.
- 8:16SPEAKER_01: And
- 8:17Some of these results and some of these write-ups that I'm
- 8:19doing are premised on
- 8:20SPEAKER_01: the idea that the data is actually correct,
- 8:23SPEAKER_01: but you're not doing analysis of contemporary data.
- 8:28You're doing analysis of DNA,
- 8:31of genome sequences,
- 8:33or doing the SNP capture array.
- 8:36SPEAKER_01: Thousands of, sometimes 10,000 years old.
- 8:38You know,
- 8:38I think this is 14,000 years old,
- 8:40this red deer cave sample that you're talking about.
- 8:42SPEAKER_01: Can you tell the listener,
- 8:44SPEAKER_01: you know,
- 8:45SPEAKER_01: in ways they can understand how you figure out that this is,
- 8:49SPEAKER_01: one,
- 8:49SPEAKER_01: the data from the individual in question,
- 8:52SPEAKER_01: and two,
- 8:53SPEAKER_01: I'm assuming,
- 8:54SPEAKER_01: I mean,
- 8:54SPEAKER_01: I know,
- 8:55SPEAKER_01: there is,
- 8:56SPEAKER_01: you know,
- 8:56SPEAKER_01: damage and degradation.
- 8:58SPEAKER_01: How you clean that up so you can compare it to modern genome samples?
- 9:03SPEAKER_00: Yeah, those are great questions.
- 9:08SPEAKER_00: I'll say,
- 9:08SPEAKER_00: kind of as a side,
- 9:09SPEAKER_00: MZR as a process,
- 9:11the project actually was a brutal awakening for me that the
- 9:16data in general should
- 9:17SPEAKER_00: not be blindly trusted.
- 9:19SPEAKER_00: I started it my first year, and it was sort of the...
- 9:24SPEAKER_00: You know, some light in my eyes was replaced by dark cynicism.
- 9:29SPEAKER_00: But no, it's true.
- 9:30SPEAKER_00: It's like people do their best.
- 9:32But like,
- 9:33you know,
- 9:33it's really hard to get these DNA out of these samples.
- 9:35SPEAKER_00: It's really, really difficult.
- 9:36SPEAKER_00: And many of them fail.
- 9:37SPEAKER_00: All right.
- 9:37SPEAKER_00: So to answer your question more directly,
- 9:41SPEAKER_00: I guess the first thing you do is dating the sample,
- 9:44SPEAKER_00: which is actually,
- 9:44SPEAKER_00: I think,
- 9:45an underrated but super important thing because while
- 9:48archaeologists are very often
- 9:50SPEAKER_00: totally correct about their assessment,
- 9:52SPEAKER_00: there can be things like where the sample comes from.
- 9:55There can be things like a rodent who moves the bones around
- 9:58or,
- 10:00you know,
- 10:00like,
- 10:00like,
- 10:01just like it gets mixed up in a drawer because it was actually
- 10:03excavated in the 1700s.
- 10:04And,
- 10:05you know,
- 10:05at some point in the last 300 years,
- 10:06it got,
- 10:07you know,
- 10:07mislabeled or something.
- 10:08SPEAKER_00: A kind of interesting example, which I'm not sure many people know about.
- 10:12SPEAKER_00: I'm curious if you know about it.
- 10:12SPEAKER_00: If you do, I apologize for repeating it, maybe.
- 10:15SPEAKER_00: But the Cro-Magnon Man, like the Cro-Magnon Man, like the original one,
- 10:19SPEAKER_00: at least part of its skull,
- 10:21it seems like its skull is a little bit of a mishmash of some
- 10:23very ancient person
- 10:25SPEAKER_00: and some more recent people.
- 10:27SPEAKER_00: Potentially someone was like, oh, there's a bone missing.
- 10:29SPEAKER_00: It's this bone, and kind of filled it in.
- 10:30SPEAKER_00: I don't know the details.
- 10:32SPEAKER_00: Anyway, that bone actually was dated.
- 10:35It's the only bone of the chromagnum we got DNA from because
- 10:38it's the most recent
- 10:38SPEAKER_00: bone,
- 10:39SPEAKER_00: and it's medieval,
- 10:40SPEAKER_00: basically.
- 10:41SPEAKER_00: I think that's published in Current Biology, some paper.
- 10:44SPEAKER_00: It's buried in a supplement.
- 10:45SPEAKER_00: Um,
- 10:46SPEAKER_00: you know,
- 10:46SPEAKER_00: it's not particularly interesting,
- 10:47SPEAKER_00: but it's a little bit sad that like the Cro-Magnon man,
- 10:50SPEAKER_00: which like Cro-Magnon,
- 10:51SPEAKER_00: you know,
- 10:51SPEAKER_00: that's like the name given for,
- 10:53SPEAKER_00: you know,
- 10:53SPEAKER_00: cavemen for,
- 10:54SPEAKER_00: you know,
- 10:55SPEAKER_00: it's sort of,
- 10:55SPEAKER_00: uh,
- 10:56SPEAKER_00: not actually maybe even from that period.
- 10:58SPEAKER_00: Uh, yeah.
- 10:58So the first thing I do is dating usually with carbon dating,
- 11:00but there's other methods.
- 11:01SPEAKER_00: Um, and once you have the data, um, uh,
- 11:05SPEAKER_00: you can look at a bunch of metrics.
- 11:08SPEAKER_00: For example,
- 11:09SPEAKER_00: the amount of damage just in the entire library,
- 11:15SPEAKER_00: which the library,
- 11:15SPEAKER_00: I should clarify,
- 11:16SPEAKER_00: maybe is like the BAM you get from the data.
- 11:21SPEAKER_00: It's actually like a physical thing,
- 11:22but you make some physical kind of soup of DNA and kind of
- 11:27biochemically in ways
- 11:28SPEAKER_00: I'm going to kind of just ignore.
- 11:30And then you sequence it and you get this FASC files you're
- 11:33mentioning and you
- 11:33SPEAKER_00: align to get the BAM.
- 11:34SPEAKER_00: So once you have that, you can look at the...
- 11:37rate of mismatching to like the human reference genome and of
- 11:41course you know we're
- 11:41not all clones there'll be mismatches all over the place
- 11:44although actually you know
- 11:45the amount of human differences per base is like one in a
- 11:48thousand so they're not
- 11:49SPEAKER_00: that common um but you want to see that there's an excess of c to t um
- 11:56SPEAKER_00: kind of almost like mutations, not mutations.
- 11:58SPEAKER_00: They're like implicitly not.
- 11:59They're kind of errors in some way or like modifications that
- 12:02happen to the DNA through time.
- 12:04SPEAKER_01: Yeah, and just to be clear, we have A, C, T, and G for the DNA.
- 12:08SPEAKER_01: That's what you're talking about.
- 12:10SPEAKER_00: Yeah, sorry.
- 12:11SPEAKER_00: There's a process called deamination,
- 12:14SPEAKER_00: which I believe,
- 12:15SPEAKER_00: I'm not a biochemist,
- 12:16SPEAKER_00: I'm really a fake biologist,
- 12:17SPEAKER_00: my background is math and computer science,
- 12:19SPEAKER_00: that there's this aminase that gets removed from the cytosine.
- 12:24SPEAKER_00: It actually turns it into a uracel,
- 12:26SPEAKER_00: which people may not be familiar with,
- 12:27SPEAKER_00: is a U,
- 12:29SPEAKER_00: and it's the RNA version of a T.
- 12:33SPEAKER_00: I think that's the main difference,
- 12:34SPEAKER_00: or maybe some other differences.
- 12:35SPEAKER_00: Again, I'm not actually a real biologist, but RNA has a U instead of a T.
- 12:40And so when you read it with a DNA sequencer,
- 12:43it reads as a T,
- 12:43because it's quite similar.
- 12:46SPEAKER_00: And so these cytosines,
- 12:48SPEAKER_00: the Cs,
- 12:48SPEAKER_00: get turned into really uracils,
- 12:51SPEAKER_00: which is the U,
- 12:52SPEAKER_00: but it's read as a T.
- 12:54SPEAKER_00: And you can look at the damage rate overall,
- 12:57SPEAKER_00: and that should,
- 12:58SPEAKER_00: in theory,
- 12:59SPEAKER_00: be elevated compared to modern DNA.
- 13:02SPEAKER_00: And, like, it should just basically be, again, C to T damage.
- 13:05SPEAKER_00: And then,
- 13:05SPEAKER_00: additionally,
- 13:05you want to see it on the ends of the reads because the ends of
- 13:09the reads sort of
- 13:09SPEAKER_00: get frayed.
- 13:10SPEAKER_00: You can imagine they're kind of, like, getting broken apart.
- 13:13SPEAKER_00: And so you should look at the ends of the reads.
- 13:15SPEAKER_00: And if you have what's called a single-stranded library,
- 13:17SPEAKER_00: which I don't want to get too technical here,
- 13:18SPEAKER_00: but basically,
- 13:19SPEAKER_00: and also this is not my expertise either,
- 13:22SPEAKER_00: basically,
- 13:23SPEAKER_00: if you read in,
- 13:24SPEAKER_00: like...
- 13:25SPEAKER_00: sort of just like the one half of the DNA.
- 13:28You'll get the C to T on both ends because they're kind of
- 13:30afraid on both ends.
- 13:31SPEAKER_00: But if you do what's called double-stranded library prep,
- 13:33where you actually bond it to a kind of novel DNA on the other
- 13:35side and kind of
- 13:36SPEAKER_00: make like a mirror copy.
- 13:37SPEAKER_00: Then on the other end,
- 13:39SPEAKER_00: on what's called the five prime end,
- 13:41SPEAKER_00: I believe,
- 13:41SPEAKER_00: you'll still get the C to T damage.
- 13:43SPEAKER_00: You actually get the reverse on the three prime end,
- 13:47SPEAKER_00: which is G to A,
- 13:48SPEAKER_00: because G in DNA is always kind of like attached to a...
- 13:53SPEAKER_00: C and T is always attached to an A.
- 13:56SPEAKER_00: So the C's turn into T's,
- 13:57SPEAKER_00: but that actually just makes G's turn into A's.
- 13:59SPEAKER_00: I hope that's not too technical.
- 14:01SPEAKER_00: And that's kind of the...
- 14:04first thing you can do and there's other things as well to look
- 14:06at contamination um
- 14:07which i can um i gotta i like to talk about like the like
- 14:11looking at the mt and the
- 14:12x chromosome but i'll i'll wait there in case you want to jump
- 14:15in no no i mean so
- 14:16you know you were saying you're a fake biologist which uh you
- 14:19know some of the
- 14:19listeners would probably like be like a little confused or
- 14:21laughing at that this is
- 14:22SPEAKER_01: not an uncommon thing um i will tell you a story um
- 14:26SPEAKER_01: Do you know Michael Torelli at UC Davis?
- 14:29SPEAKER_01: Quantitative.
- 14:30SPEAKER_01: Sounds familiar, but I like that.
- 14:33SPEAKER_01: Anyway, so Torelli, one, his writing is illegible.
- 14:40SPEAKER_01: He's famous for that.
- 14:40SPEAKER_01: He was Felsenstein's postdoc.
- 14:43SPEAKER_01: And I remember one time,
- 14:50I was in a seminar with Torelli and he was explaining about
- 14:56recombination rates and
- 14:59how they decrease at the telomeres and they're maximized at the
- 15:02centromeres.
- 15:05And all of us who had taken the recent molecular genetics
- 15:09course looked at each
- 15:11SPEAKER_01: other,
- 15:12SPEAKER_01: because in humans,
- 15:13SPEAKER_01: it's the opposite.
- 15:15SPEAKER_01: And so,
- 15:17SPEAKER_01: Jarelli,
- 15:17SPEAKER_01: he's the National Academy of Sciences,
- 15:19SPEAKER_01: he's a very,
- 15:19SPEAKER_01: very prominent quantitative geneticist,
- 15:21SPEAKER_01: but biology is a massive field,
- 15:25SPEAKER_01: and no one is going to be the specialist,
- 15:26SPEAKER_01: or no one's going to know...
- 15:29SPEAKER_01: No one's going to know very much of it, to be candid, right?
- 15:33SPEAKER_00: Yeah, no, it's true.
- 15:33SPEAKER_00: I think it's also true of other...
- 15:36The well of knowledge is so deep that if you want to become
- 15:39specialized in knock
- 15:40SPEAKER_00: theory or something in some obscure field of mathematics,
- 15:43SPEAKER_00: you're going to not know much about...
- 15:44SPEAKER_00: I mean,
- 15:45SPEAKER_00: you'll know much more than the average person on the street,
- 15:47SPEAKER_00: but you won't know much compared to your fellow experts at other places.
- 15:51SPEAKER_00: Yeah.
- 15:52SPEAKER_00: I just like our field specifically.
- 15:54SPEAKER_00: I mean, you might relate.
- 15:55SPEAKER_00: Maybe you don't.
- 15:56SPEAKER_00: But I sort of feel like a jack-of-all-trades, like a master of none.
- 15:59SPEAKER_00: I potentially...
- 16:01know far more math and computer science than some other people
- 16:05in like who are
- 16:06doing like real like biology but then like I know so much less
- 16:09than like people in
- 16:10math and computer science programs and of course I know so much
- 16:13more biology than
- 16:14like my friends who are doing math and computer science but
- 16:15like so much less
- 16:16SPEAKER_00: computer science and math than them.
- 16:19SPEAKER_01: Well,
- 16:19SPEAKER_01: I mean,
- 16:19SPEAKER_01: so one thing that I will say,
- 16:21SPEAKER_01: you know,
- 16:21SPEAKER_01: in discussions,
- 16:22SPEAKER_01: this meta question here,
- 16:23I think part of the things in genomics is that the first human
- 16:26genome was,
- 16:27SPEAKER_01: the draft was completed 25 years ago,
- 16:30SPEAKER_01: probably around when you were born,
- 16:31SPEAKER_01: right?
- 16:32SPEAKER_01: So it's a very, very young field, and it's moving extremely fast.
- 16:37SPEAKER_01: And so,
- 16:38SPEAKER_01: you know,
- 16:38SPEAKER_01: you could be really good at doing something,
- 16:41SPEAKER_01: and then it's totally irrelevant,
- 16:45SPEAKER_01: right?
- 16:46SPEAKER_01: I mean,
- 16:48SPEAKER_01: you know,
- 16:49SPEAKER_01: I think it's,
- 16:50SPEAKER_01: I've been out of academia for a little while,
- 16:51SPEAKER_01: you know,
- 16:52SPEAKER_01: with RNA-seq and,
- 16:53SPEAKER_01: you know,
- 16:54SPEAKER_01: understanding gene expression with RNA,
- 16:57SPEAKER_01: you know,
- 16:57SPEAKER_01: sequencing,
- 16:58SPEAKER_01: for example,
- 16:59SPEAKER_01: I know when I was in grad school,
- 17:00SPEAKER_01: some of the younger grad students,
- 17:01SPEAKER_01: they didn't know what gene expression arrays were.
- 17:03SPEAKER_01: Do you know what they are?
- 17:05SPEAKER_01: Uh, only vaguely.
- 17:07SPEAKER_01: Okay.
- 17:07SPEAKER_01: Well,
- 17:07SPEAKER_01: they were again,
- 17:09SPEAKER_01: like older biologists are going to be laughing because they were very,
- 17:12SPEAKER_01: very cutting edge in 2005.
- 17:13SPEAKER_01: Okay.
- 17:15But they just kind of got phased out about five years later
- 17:19when whole genome
- 17:20SPEAKER_01: sequencing came in.
- 17:20So gene expression arrays just like were the way you figured
- 17:23out like which genes
- 17:24SPEAKER_01: were being expressed in what tissues and stuff.
- 17:26SPEAKER_01: They didn't do RNA sequencing.
- 17:27SPEAKER_01: Right.
- 17:28SPEAKER_01: Um,
- 17:28And so there were people,
- 17:30they knew how to read these arrays,
- 17:31how to analyze these arrays.
- 17:32SPEAKER_01: There were whole pipelines for the arrays to understand them,
- 17:37SPEAKER_01: and it was just totally irrelevant within five years.
- 17:40And so I think genomics is special that way in terms of you're
- 17:44always having to
- 17:44update and you're always having to accept that you don't
- 17:47totally understand what's
- 17:48SPEAKER_01: going on.
- 17:49SPEAKER_01: And we're talking about contamination here.
- 17:51I'll give you a quick story that I've heard many times from
- 17:53many different people.
- 17:54SPEAKER_01: But initially,
- 17:56geneticists were not aware of artifacts introduced by the
- 18:01sequencing technology you used.
- 18:04SPEAKER_01: And so there was a period where,
- 18:06SPEAKER_01: I don't think it was ever published,
- 18:08SPEAKER_01: but I think some things like this have been published.
- 18:10But basically,
- 18:12someone created a phylogeny of a population,
- 18:16so it's within a single species.
- 18:19SPEAKER_01: And they found this novel,
- 18:20SPEAKER_01: crazy structure that they had never expected,
- 18:22SPEAKER_01: like relationships that were never expected.
- 18:25SPEAKER_01: Right before publication,
- 18:26SPEAKER_01: someone looked at the methods,
- 18:28SPEAKER_01: and they matched the IDs of the individuals to the sequencing platform,
- 18:32because this was in a period where the sequencing was extremely
- 18:35expensive,
- 18:36SPEAKER_01: it was very bespoke,
- 18:37SPEAKER_01: and not many labs really had access to cores,
- 18:40and it turned out the phylogeny was just a recapitulation of
- 18:43the sequencing
- 18:43SPEAKER_01: technology.
- 18:44SPEAKER_01: That's funny.
- 18:46Because the errors within each sequencing technology were
- 18:51different from the errors
- 18:52SPEAKER_01: in other sequencing technology,
- 18:53SPEAKER_01: and so it was a phylogeny of errors,
- 18:55SPEAKER_01: right?
- 18:56SPEAKER_01: And so obviously today that wouldn't happen,
- 18:59SPEAKER_01: partly because sequencing is so cheap and it's so standardized and so,
- 19:02SPEAKER_01: you know,
- 19:02SPEAKER_01: but,
- 19:03SPEAKER_01: you know,
- 19:04SPEAKER_01: 10, 15 years ago, this was still happening, right?
- 19:07SPEAKER_00: All the ways that,
- 19:09SPEAKER_00: you know,
- 19:09SPEAKER_00: I mean,
- 19:10SPEAKER_00: I guess,
- 19:10SPEAKER_00: like I said before,
- 19:11SPEAKER_00: you don't trust the data fully.
- 19:14But,
- 19:14yeah,
- 19:14no,
- 19:14I mean,
- 19:15I think to some degree there still is,
- 19:16like,
- 19:16evidence of these things.
- 19:18SPEAKER_00: I know that someone else in the lab,
- 19:19SPEAKER_00: a postdoc named Roman Ferrier,
- 19:22SPEAKER_00: just published a preprint on,
- 19:24SPEAKER_00: like,
- 19:25SPEAKER_00: a set of SNPs that are more...
- 19:27SPEAKER_00: like,
- 19:27SPEAKER_00: compatible across,
- 19:28SPEAKER_00: like,
- 19:29SPEAKER_00: 1240K,
- 19:29SPEAKER_00: Twist,
- 19:29SPEAKER_00: and Shotgun,
- 19:30SPEAKER_00: because it's not huge,
- 19:32SPEAKER_00: but you do get sort of these weird things where you get,
- 19:34SPEAKER_00: like,
- 19:34SPEAKER_00: individual one,
- 19:35SPEAKER_00: Shotgun,
- 19:36SPEAKER_00: individual two,
- 19:38SPEAKER_00: 1240K.
- 19:38SPEAKER_00: For people who don't know,
- 19:39SPEAKER_00: that's,
- 19:40SPEAKER_00: like,
- 19:40SPEAKER_00: a different type of,
- 19:41SPEAKER_00: it's,
- 19:41SPEAKER_00: like,
- 19:41SPEAKER_00: different types of sequencing.
- 19:43SPEAKER_00: You,
- 19:43SPEAKER_00: like,
- 19:43SPEAKER_00: compare the two individuals,
- 19:44SPEAKER_00: like,
- 19:45SPEAKER_00: you know,
- 19:45SPEAKER_00: you have two individuals,
- 19:46SPEAKER_00: the same individual,
- 19:47SPEAKER_00: but one's Shotgun and one's 1240K,
- 19:49SPEAKER_00: so that's four,
- 19:50fake individuals but it's really just two and actually you see
- 19:52an attraction
- 19:53between the different sequencing profiles and like I know also
- 19:55in our lab when you
- 19:57do what's called like variant calling which is where again you
- 19:59have all these
- 19:59different reads of DNA and you kind of choose okay what's the
- 20:01actual site like is
- 20:02SPEAKER_00: it actually a C or a T or whatever we like
- 20:06SPEAKER_00: don't do majority calling,
- 20:07SPEAKER_00: which kind of is the first thing you might think,
- 20:09SPEAKER_00: like where you look at the reads and say,
- 20:10SPEAKER_00: oh,
- 20:10SPEAKER_00: the most is C,
- 20:11SPEAKER_00: so we should choose C.
- 20:13SPEAKER_00: We actually just kind of do a random,
- 20:14SPEAKER_00: you pick a random read,
- 20:16SPEAKER_00: which is kind of bad in some way because,
- 20:17SPEAKER_00: you know,
- 20:18SPEAKER_00: if one read is erroneous and then you pick that random read,
- 20:22SPEAKER_00: you know,
- 20:22SPEAKER_00: you put one in 10 chance of getting it wrong,
- 20:23SPEAKER_00: even though really you have,
- 20:24SPEAKER_00: you know,
- 20:24SPEAKER_00: the information is strongly telling you that it's,
- 20:27SPEAKER_00: It's the other one.
- 20:28SPEAKER_00: But the reason we do that is because,
- 20:30SPEAKER_00: actually,
- 20:30SPEAKER_00: if each read is slightly biased at heterozygous sites,
- 20:33SPEAKER_00: which are sites where the individual actually has both copies,
- 20:38SPEAKER_00: as you get more reads,
- 20:38SPEAKER_00: the bias builds up.
- 20:39SPEAKER_00: So it's definitely something we think about.
- 20:41SPEAKER_00: And I mean, I don't want to say that, like,
- 20:44I don't want anyone to come away from this thinking that the
- 20:46field is fake.
- 20:46SPEAKER_00: I mean, the field is really robust.
- 20:48SPEAKER_00: It's actually incredible the insights we can make.
- 20:52But definitely,
- 20:53you know,
- 20:54it's always suspicion and skepticism are often really valuable.
- 20:58SPEAKER_01: Yeah,
- 20:59SPEAKER_01: and when you're on the cutting edge of something,
- 21:00when you're on the cutting edge of a science or a new set of
- 21:04methodologies,
- 21:05SPEAKER_01: people actually don't know often what type of errors are very common.
- 21:11SPEAKER_01: And so people learn from the mistakes that they make.
- 21:13And so papers like yours are important because the whole idea
- 21:18is other people read
- 21:18SPEAKER_01: the papers and
- 21:20SPEAKER_01: and they know to be careful of certain things,
- 21:22SPEAKER_01: and they are careful of certain things right now.
- 21:25SPEAKER_01: For example,
- 21:25SPEAKER_01: you know,
- 21:26SPEAKER_01: and some of the listeners know,
- 21:28some of the early detection of natural selection in certain
- 21:33populations turned out
- 21:35SPEAKER_01: to be just population structure artifacts.
- 21:37SPEAKER_01: And people are very, very aware about that right now.
- 21:40SPEAKER_01: They were not super aware of that a decade ago, but now they are.
- 21:44SPEAKER_01: So it's a learning process.
- 21:46SPEAKER_01: So can you talk about
- 21:48SPEAKER_01: this particular paper about the Red Deer Cave individuals,
- 21:51what the problems specifically were that you saw and how that
- 21:55impacted the
- 21:55SPEAKER_01: conclusions that you could or couldn't draw?
- 21:59SPEAKER_00: Yeah.
- 22:00So I guess the first thing I noticed just was that I didn't
- 22:05think the conclusions
- 22:05SPEAKER_00: were super well supported.
- 22:06I thought they were just sort of,
- 22:08the sample also has low coverage and a lack of power.
- 22:10SPEAKER_00: And then I started to look into the data more.
- 22:14SPEAKER_00: Just to be clear, tell them what low coverage is.
- 22:17ah it's a good good point yeah so basically if you were to look
- 22:21at the average
- 22:22SPEAKER_00: position in the genome and say how many um times do we have a read that's
- 22:27overlapping this position um you know if you were to have 30x
- 22:31coverage means on
- 22:32average there's 30 reads that overlap like 30 sequences of dna
- 22:36that overlap the uh
- 22:38SPEAKER_00: that position um
- 22:39which then gives you,
- 22:40like,
- 22:40very high confidence what the position actually is.
- 22:42SPEAKER_00: You could even,
- 22:42SPEAKER_00: like,
- 22:42SPEAKER_00: do what's called diploid calling and get both reads because,
- 22:45SPEAKER_00: you know,
- 22:45SPEAKER_00: humans have two copies,
- 22:47SPEAKER_00: one for their mom,
- 22:47SPEAKER_00: one for their dad.
- 22:49SPEAKER_00: But this individual had, like, 0.03x coverage, which is quite low.
- 22:53SPEAKER_00: I mean, there's definitely lower, but, like,
- 22:54SPEAKER_00: You know,
- 22:55SPEAKER_00: in our field with ancient DNA,
- 22:56SPEAKER_00: you often don't get 30X coverage,
- 22:57SPEAKER_00: which is maybe like a modern genome.
- 22:59SPEAKER_00: You might get like, you know, 2X coverage or like 0.5X coverage.
- 23:02SPEAKER_00: But 0.03 is like quite low.
- 23:04SPEAKER_00: And so like they were saying that some Z-scores,
- 23:06which are sort of like a measure of significance with some like
- 23:09questions they were
- 23:09SPEAKER_00: asking were not significant,
- 23:11SPEAKER_00: which is interesting.
- 23:12SPEAKER_00: But like it's not clear.
- 23:13SPEAKER_00: It wasn't clear to me that this was because...
- 23:19SPEAKER_00: the answer was neutral.
- 23:21SPEAKER_00: And the question they were asking is, is MZR closer to publishing A or B?
- 23:26SPEAKER_00: And they said, well, we can't tell.
- 23:27And is it because that they're equally close to both,
- 23:29or is it because of a lack of coverage?
- 23:31SPEAKER_00: So then I started going into the
- 23:33bam more carefully and there's like a number of weird things
- 23:36about it like i
- 23:37actually don't remember the order but like one thing is it's
- 23:39clearly a mixture of
- 23:40shotgun and 1240k sequences um and it's it's not clearly
- 23:44separated off and then
- 23:46even though i then look for contamination because it looks it
- 23:48looks very much like
- 23:49i'm like the genome is very closely related to or has a lot of
- 23:53affinity towards i
- 23:53shouldn't say related to but has a lot of affinity towards
- 23:55modern han chinese which
- 23:57SPEAKER_00: is like
- 23:57that's like what you expect for contamination,
- 23:59given that this was a Chinese lab.
- 24:00SPEAKER_00: In the same way,
- 24:01you might expect Northern European contamination from like a
- 24:04lab in Germany or
- 24:05SPEAKER_00: like,
- 24:05SPEAKER_00: you know,
- 24:05SPEAKER_00: a sample from Germany,
- 24:06SPEAKER_00: which may be a German archaeologist,
- 24:08SPEAKER_00: you know,
- 24:09SPEAKER_00: dug up or something.
- 24:10SPEAKER_00: So I thought maybe it was contaminated.
- 24:12SPEAKER_00: But actually, it does show the typical pattern of...
- 24:17like damage which you should expect which actually um was a
- 24:21little bit surprising
- 24:22to me and then you know actually it's very interesting and that
- 24:24was the claim was
- 24:24true but i kept looking at it and actually the pattern of
- 24:28damage which of like c to
- 24:30t damage at the end of the read um was also present for like c
- 24:34to g or c to a or a
- 24:37SPEAKER_00: to g or a to t or anything else like all of the um
- 24:42sort of subcategories of like possible mutations were all
- 24:45extremely enriched at the
- 24:47SPEAKER_00: ends of the reads,
- 24:47SPEAKER_00: which is not what you expect.
- 24:50And it seems to indicate that some other process was
- 24:53potentially generating that
- 24:56SPEAKER_00: CDT profile,
- 24:58SPEAKER_00: which the authors had looked at and said,
- 25:00SPEAKER_00: oh,
- 25:00SPEAKER_00: well,
- 25:00SPEAKER_00: it doesn't look contaminated.