Latest / Razib Khan's Unsupervised Learning / Jonathan Beauchamp: Genoeconomics arise!
Transcript
- 0:00SPEAKER_02: This podcast is brought to you by the Albany Public Library Main Branch and
- 0:03SPEAKER_02: the generosity of listeners like you.
- 0:09SPEAKER_02: What is a podcast?
- 0:16SPEAKER_02: God, Daddy, these people talk as much as
- 0:24you do.
- 0:26SPEAKER_02: Razib Khan's Unsupervised Learning.
- 0:29SPEAKER_01: Hey everybody, this is Razib Khan with the Unsupervised Learning Podcast.
- 0:32SPEAKER_01: Today, I am with my friend Jonathan Beauchamp,
- 0:38SPEAKER_01: who is an associate professor at George Mason University in Virginia,
- 0:44SPEAKER_01: and his work is in behavioral economics,
- 0:46SPEAKER_01: applied microeconomics, and genome economics.
- 0:51SPEAKER_01: And so I actually went in reverse order from what's listed on the faculty page.
- 0:56SPEAKER_01: I think most people know what behavioral economics is and applied microeconomics.
- 1:01SPEAKER_01: Obviously, people know what that is.
- 1:03SPEAKER_01: But genome economics, I think you could guess just by the structure of the name what it is.
- 1:08SPEAKER_01: But most people probably have not heard about it.
- 1:12SPEAKER_01: So, yeah, Jonathan, could you just start telling my audience what genome economics is?
- 1:18SPEAKER_00: Yeah, of course.
- 1:19SPEAKER_00: So first, well, Razib, thanks for inviting me.
- 1:21SPEAKER_00: I'm very happy to be here.
- 1:22SPEAKER_00: It's a pleasure.
- 1:23SPEAKER_00: So to your question,
- 1:25SPEAKER_00: so genome economics is a term we,
- 1:27SPEAKER_00: well, some of my co-authors have coined, that's basically genetics and economics.
- 1:34SPEAKER_00: And originally,
- 1:35SPEAKER_00: this was really a bit,
- 1:37SPEAKER_00: you can think of it as the medical genetics of economic traits.
- 1:41SPEAKER_00: So just as there are genetic studies,
- 1:43SPEAKER_00: like large-scale genome study GWASs,
- 1:45SPEAKER_00: most of your audience will know that,
- 1:48SPEAKER_00: We wanted to do the same to study traits of interest to social scientists
- 1:55SPEAKER_00: and economists,
- 1:56SPEAKER_00: such as risk preferences,
- 1:58SPEAKER_00: educational attainment, income, subjective well-being, etc.
- 2:02SPEAKER_00: So that was the first kind of big wave of work.
- 2:06SPEAKER_00: Well, actually,
- 2:06SPEAKER_00: there was another wave before that that's basically demonstrating that much of
- 2:10SPEAKER_00: the research at that time was junk, but you know, it's Back to your question.
- 2:14SPEAKER_00: So the other segment of junior economics is really about,
- 2:19SPEAKER_00: in my opinion,
- 2:19SPEAKER_00: it's about applying this in economics and,
- 2:24SPEAKER_00: you know,
- 2:24SPEAKER_00: studying,
- 2:25SPEAKER_00: you know,
- 2:26SPEAKER_00: The role of genetics and environment and human capital formation,
- 2:29SPEAKER_00: production function, that's economic jargon, just to say human development really.
- 2:34SPEAKER_00: Intergenerational mobility and equality and so on, nature and nurture.
- 2:38SPEAKER_00: And, you know,
- 2:40SPEAKER_00: I define this pretty broadly,
- 2:41SPEAKER_00: but anything where there's intersection between applied economics,
- 2:44SPEAKER_00: you know,
- 2:44SPEAKER_00: tackling questions of,
- 2:45SPEAKER_00: you know, questions of traditional interest to economists and genetics, I think.
- 2:52SPEAKER_00: Yes, that would be, that's my definition.
- 2:54SPEAKER_01: I mean, I think one way you could say it is like,
- 2:55SPEAKER_01: you know,
- 2:56SPEAKER_01: okay, the questions are economic and the methods integrate genetics, right?
- 3:01SPEAKER_00: Correct.
- 3:01SPEAKER_00: Exactly.
- 3:01SPEAKER_00: That's a good, that's a shorter definition, definitely.
- 3:06SPEAKER_01: Well, you know,
- 3:07SPEAKER_01: you're a professor,
- 3:08SPEAKER_01: so you're going to profess a little bit and like speak at length.
- 3:11SPEAKER_01: So, you know,
- 3:12SPEAKER_01: there's behavioral economics,
- 3:15SPEAKER_01: which I think applies,
- 3:15SPEAKER_01: you know,
- 3:16SPEAKER_01: psychology, cognitive science, other things, so, you know, it kind of makes sense.
- 3:21SPEAKER_01: Excuse me, how many Gino economists are there right now?
- 3:25SPEAKER_00: That's a good question.
- 3:26SPEAKER_00: It depends how you count.
- 3:27SPEAKER_00: I would say roughly...
- 3:31SPEAKER_00: You know, a hundred.
- 3:32SPEAKER_00: But it depends how you count.
- 3:34SPEAKER_00: I would say social science geneticists is probably the broader community in that field.
- 3:39SPEAKER_00: And I include some political scientists,
- 3:41SPEAKER_00: some sociologists, some gene economists, and other scholars that are adjacent.
- 3:47SPEAKER_00: I'd say, you know, a hundred maybe.
- 3:49SPEAKER_00: And then if you broaden to people who have published something on the topic
- 3:53SPEAKER_00: of genetics and social science, that's much broader.
- 3:55SPEAKER_00: But I think people are active in this.
- 3:57SPEAKER_00: If you look at the participation,
- 3:58SPEAKER_00: the conferences,
- 3:59SPEAKER_00: the main conferences in the field,
- 4:01SPEAKER_00: There's one in Europe and roughly two in the US that occur every year.
- 4:06SPEAKER_00: I'd say, you know, it would reach to, well, you know, maybe 50 each.
- 4:12SPEAKER_00: So, you know, you can really make the case of about 100 of us, roughly.
- 4:16SPEAKER_01: Well,
- 4:17SPEAKER_01: so there's a hundred of you and you've been involved in educational attainment work
- 4:21SPEAKER_01: and other things like that.
- 4:22SPEAKER_01: Can you talk about that a little bit?
- 4:25SPEAKER_01: Some of my listeners will know what educational attainment, what that is.
- 4:29SPEAKER_01: We talked to James Lee and Alex Young, et cetera, a lot.
- 4:33SPEAKER_01: Talk about why, as an economist, you're interested in it.
- 4:37SPEAKER_01: Yeah, just talk about it.
- 4:40SPEAKER_00: Education attainment is a very important variable in labor economics,
- 4:45SPEAKER_00: in growth economics, and everything.
- 4:50SPEAKER_00: Public health economics, everything.
- 4:51SPEAKER_00: It's basically correlated with everything That one,
- 4:56SPEAKER_00: you know, almost everything that one cares about.
- 4:59SPEAKER_00: And I should define years of education attainment operationally,
- 5:04SPEAKER_00: we define it as just years of education.
- 5:05SPEAKER_00: So, you know,
- 5:05SPEAKER_00: if you've got 12 degrees,
- 5:07SPEAKER_00: 12 years of education,
- 5:09SPEAKER_00: Of schooling until you completed high school and then stopped, then you're 12.
- 5:12SPEAKER_00: If you've got an undergrad on top of this, that's 16.
- 5:14SPEAKER_00: If you've got a master's, then that's probably 17, et cetera.
- 5:20SPEAKER_00: So it's just very correlated with everything,
- 5:23SPEAKER_00: ranging from smoking, health, longevity, income, unemployment.
- 5:29SPEAKER_00: It's a very predictive variable.
- 5:32SPEAKER_00: Now, the reason for that is...
- 5:36SPEAKER_00: Still debated.
- 5:37SPEAKER_00: There's lots of, there's a whole, an enormous literature on the treatment effect of education.
- 5:42SPEAKER_00: And, you know, it finds typically pretty sizable positive effects.
- 5:45SPEAKER_00: There's still debates as to whether how much of this is due to signaling versus human capital.
- 5:51SPEAKER_00: That is to say,
- 5:52SPEAKER_00: when you get one,
- 5:53SPEAKER_00: let's say, college degree, you know, you'll get more income and so on.
- 5:56SPEAKER_00: How much of that is due to just,
- 5:57SPEAKER_00: you know,
- 5:58SPEAKER_00: the sheepskin effect of adding a degree versus what you've learned,
- 6:02SPEAKER_00: the human capital you've acquired by studying for it.
- 6:05SPEAKER_00: Um, so, so basically we did, so that's education attainment.
- 6:09SPEAKER_00: And then we did the first large scale GWAS education attainment.
- 6:12SPEAKER_00: And then me and my co-authors, we did four of them actually.
- 6:15SPEAKER_00: Um, yeah, I can keep going on, on this if you want.
- 6:19SPEAKER_01: Yeah, you know,
- 6:20SPEAKER_01: can you actually talk about like,
- 6:21SPEAKER_01: you know,
- 6:21SPEAKER_01: you kind of alluded to it,
- 6:22SPEAKER_01: but like educational attainment,
- 6:25SPEAKER_01: you know,
- 6:25SPEAKER_01: why you're using educational attainment instead of IQ and stuff like that and
- 6:29SPEAKER_01: the pros and cons of it, right?
- 6:30SPEAKER_00: Yeah, that's very good.
- 6:31SPEAKER_00: So ultimately, it was mainly the way we started on this is just by going for low-hanging fruits.
- 6:37SPEAKER_00: We were,
- 6:38SPEAKER_00: when I started in this with David Cesarini,
- 6:40SPEAKER_00: who's at NYU, and Dan Benjamin also, I was in touch with him.
- 6:44SPEAKER_00: The initial project was me and David.
- 6:48SPEAKER_00: Then we worked on many other things along the way.
- 6:51SPEAKER_00: We had no budgets.
- 6:53SPEAKER_00: We were PhD students in Cambridge, Massachusetts.
- 6:56SPEAKER_00: And we...
- 6:58SPEAKER_00: Yeah, we were very interested into this.
- 6:59SPEAKER_00: We saw a lot of claims of,
- 7:01SPEAKER_00: you know,
- 7:01SPEAKER_00: genes for entrepreneurship,
- 7:02SPEAKER_00: genes for happiness,
- 7:04SPEAKER_00: gene for risk-taking,
- 7:05SPEAKER_00: gene for, like,
- 7:06SPEAKER_00: the MAOA receptor,
- 7:08SPEAKER_00: the DRD4, DRD2 dopamine receptor genes, and whatever trait you want.
- 7:12SPEAKER_00: There was all these claims, so we were very skeptical of these, so we...
- 7:17SPEAKER_00: Our first contribution was to show that lots of this was junk, basically.
- 7:20SPEAKER_00: But then we wanted to do something more constructive and try to do it the right way.
- 7:26SPEAKER_00: This was in the late 2000s, like just 2009, 2010.
- 7:29SPEAKER_00: We were aware of all these advances in medical genetics.
- 7:32SPEAKER_00: Using GWAS, genome-wide association studies.
- 7:35SPEAKER_00: And we said, why don't we do this for an economic trait?
- 7:37SPEAKER_00: And obviously the, you know, or a trait of interest social scientists.
- 7:41SPEAKER_00: So IQ is obviously an important one.
- 7:43SPEAKER_00: But it's hard to get at that stage.
- 7:44SPEAKER_00: And still, you know, now less so.
- 7:46SPEAKER_00: But, you know, it was hard to get large-scale data sets to study most traits.
- 7:50SPEAKER_00: But education was typically collected as a covariate.
- 7:54SPEAKER_00: In most data sets, like, you know, so we combined various traits.
- 7:58SPEAKER_00: For the first thing we did was in the Framingham Heart Study,
- 8:01SPEAKER_00: and, you know, it's a medical, famous medical study.
- 8:04SPEAKER_00: And they collected education,
- 8:06SPEAKER_00: I think,
- 8:06SPEAKER_00: and just as one of many couverts they collected,
- 8:08SPEAKER_00: but that's So basically, it was really just a matter of opportunity.
- 8:11SPEAKER_00: So basically, it was cheap to do a GWAS on this trade, and it's a very important trade.
- 8:15SPEAKER_00: So we said, let's do it.
- 8:29SPEAKER_01: One thing I was curious about,
- 8:31SPEAKER_01: obviously educational entertainment,
- 8:32SPEAKER_01: even in terms of university and all that stuff,
- 8:35SPEAKER_01: it varies by nation in terms of accessibility,
- 8:40SPEAKER_01: affordability, all these sorts of things.
- 8:42SPEAKER_01: Actually, I haven't read all the papers on educational entertainment.
- 8:46SPEAKER_01: I've read the main ones, but not all of the ones around it.
- 8:49SPEAKER_01: There's many groups on it.
- 8:50SPEAKER_01: How do you guys talk about that methodologically?
- 8:52SPEAKER_01: Yeah.
- 8:53SPEAKER_00: So basically,
- 8:55SPEAKER_00: we do within European,
- 8:56SPEAKER_00: so far what we've been doing,
- 8:58SPEAKER_00: the four big large-scale studies, were within European ancestry people.
- 9:05SPEAKER_00: And we do this for several reasons.
- 9:07SPEAKER_00: First and primaries.
- 9:07SPEAKER_00: It's because of the issue of population stratification.
- 9:10SPEAKER_00: And once you start combining too broad a set of ancestries,
- 9:15SPEAKER_00: then there's this risk of confound that the association you'll find
- 9:19SPEAKER_00: between education attainment and a given SNP or single nucleotide polymorphism,
- 9:24SPEAKER_00: these are just genetic variants that we study,
- 9:27SPEAKER_00: Instead of reflecting a causal effect of the SNP,
- 9:30SPEAKER_00: it's going to reflect probably some confounds.
- 9:33SPEAKER_00: So basically,
- 9:34SPEAKER_00: you know,
- 9:34SPEAKER_00: the famous chopstick example,
- 9:36SPEAKER_00: if you look for genes for chopstick,
- 9:38SPEAKER_00: the SNPs,
- 9:41SPEAKER_00: and you do a So if you have a GWAS that combines Chinese from China with,
- 9:45SPEAKER_00: you know,
- 9:45SPEAKER_00: European as three Americans from the US,
- 9:47SPEAKER_00: then you'll find every SNP that,
- 9:50SPEAKER_00: every eugenic variant that varies between the average American and
- 9:54SPEAKER_00: the average Chinese to be correlated with chopstick use.
- 9:57SPEAKER_00: It's due to culture being correlated with that.
- 9:59SPEAKER_00: So basically, we just limit it to European ancestry participants.
- 10:03SPEAKER_00: Now, more modern methods allow to do cross-ancestry gewasses.
- 10:09SPEAKER_00: And I've not taken part in these so far in the last few years,
- 10:12SPEAKER_00: but this is a pretty active area now.
- 10:15SPEAKER_00: Yeah.
- 10:20SPEAKER_01: I know that there's a limit in terms of how many people you could genotype over time
- 10:27SPEAKER_01: and the samples you get and stuff like that.
- 10:29SPEAKER_01: 23andMe obviously has these huge datasets now just because they do direct-to-consumer.
- 10:34SPEAKER_01: Do you think we're going to get to a billion people in educational attainment at some point?
- 10:40SPEAKER_01: Because we've got millions.
- 10:43SPEAKER_01: Tens of millions seems plausible.
- 10:45SPEAKER_01: This is something you could collect on everyone as a characteristic, potentially.
- 10:50SPEAKER_01: What do you think about that?
- 10:53SPEAKER_01: Do you guys internally?
- 10:54SPEAKER_01: Because I know you're in a bunch of I'm assuming you're in a bunch of group chats
- 10:57SPEAKER_01: and email lists and other,
- 10:58SPEAKER_01: you know,
- 10:58SPEAKER_01: you got to go to meetings where,
- 11:01SPEAKER_01: you know, you guys talk it through and some committee and all this stuff, right?
- 11:04SPEAKER_01: Like, what are you guys thinking for the future for Educating Tim?
- 11:06SPEAKER_01: It's been about, it's a decade.
- 11:08SPEAKER_01: It's over a decade now, you know?
- 11:10SPEAKER_00: Yeah, I think we'll get to 1 billion and even really the full population of
- 11:15SPEAKER_00: the earth except a few people who abstain.
- 11:17SPEAKER_00: The reason for this is medical as clinical utility of polygenic scores and genetic predictors.
- 11:23SPEAKER_00: So basically,
- 11:24SPEAKER_00: if you think like stepping away from social genetics and just medical genetics,
- 11:30SPEAKER_00: polygenic scores They're already predictive of many, many disease.
- 11:35SPEAKER_00: They predict roughly a few percentage of the variation in most of these,
- 11:39SPEAKER_00: or many of these, rather.
- 11:41SPEAKER_00: But it's entirely a function of the sample size.
- 11:43SPEAKER_00: And so these will become more and more predictive,
- 11:46SPEAKER_00: Way before we reach a billion people genotype.
- 11:49SPEAKER_00: But the point is,
- 11:50SPEAKER_00: the cost-benefit calculus in medical genetics
- 11:53SPEAKER_00: and clinical use will switch completely very soon.
- 11:56SPEAKER_00: In fact, I would argue it's already switched.
- 11:58SPEAKER_00: Collecting,
- 11:59SPEAKER_00: genotyping someone,
- 12:00SPEAKER_00: collecting the SNPs, like let's say a 3 million SNP on someone, is so cheap.
- 12:06SPEAKER_00: I don't know the latest number.
- 12:07SPEAKER_00: Was it $20, $30?
- 12:10SPEAKER_00: Yeah, that's what
- 12:11SPEAKER_01: I heard.
- 12:11SPEAKER_01: But I think it might even be less than that because I haven't asked in like three or four years.
- 12:14SPEAKER_01: Well, actually,
- 12:15SPEAKER_01: at some point,
- 12:16SPEAKER_01: it's just going to go against the asymptote of like, okay, it's cheap enough.
- 12:20SPEAKER_01: They're not going to go to $1 even though it's like, you know what I'm saying?
- 12:22SPEAKER_00: Yeah, exactly.
- 12:23SPEAKER_00: The cost is not so much technology,
- 12:24SPEAKER_00: just like the whole setup to maintain a website and maintain an organization
- 12:28SPEAKER_00: to deliver and so on.
- 12:29SPEAKER_00: Yes.
- 12:30SPEAKER_00: So it's going to be so cheap and it's already so cheap.
- 12:33SPEAKER_00: And, you know, you think what's the cost of a blood test?
- 12:35SPEAKER_00: You know, it's It's like $100 or something, depending on who provides it and so on.
- 12:43SPEAKER_00: And that's just predictive of a bunch of current conditions about you,
- 12:46SPEAKER_00: where yes,
- 12:46SPEAKER_00: you get one genetic test at birth or whenever,
- 12:50SPEAKER_00: and it predicts all these things for your life.
- 12:54SPEAKER_00: To me, it's a no-brainer that this will become integrated in clinical practice sooner or later.
- 12:59SPEAKER_00: I mean, there's a debate actually in clinical genetics about this.
- 13:01SPEAKER_00: They say, oh, you know, polygene score is still not predictive.
- 13:03SPEAKER_00: And so why will they ever be?
- 13:05SPEAKER_00: Well, you know, because technology is improving.
- 13:08SPEAKER_00: And so at some point,
- 13:10SPEAKER_00: you know, everyone will be genotyped except people who opt out for whatever reason.
- 13:15SPEAKER_00: And this will be part of the medical file.
- 13:17SPEAKER_00: And I think, you know, it will be possible to get these data for research purposes.
- 13:22SPEAKER_00: Certainly for medical genetics research purposes, but I love it.
- 13:25SPEAKER_00: Many of them will also accept to have these data for social genomics or other research purposes.
- 13:30SPEAKER_00: So I think we're definitely getting there.
- 13:32SPEAKER_00: The question is just when.
- 13:34SPEAKER_00: And here, it's not so much a cost-benefit.
- 13:37SPEAKER_00: I used to be at the University of Toronto.
- 13:39SPEAKER_00: There's some experts there in study of medical,
- 13:43SPEAKER_00: of the genetics of autism,
- 13:44SPEAKER_00: and they've already, they genotype all their patients there.
- 13:46SPEAKER_00: So it's not because they find it very useful to diagnose a case of autism and so on.
- 13:51SPEAKER_00: So the main thing is not so much,
- 13:53SPEAKER_00: the main thing is just to get the medical practice to integrate this
- 13:57SPEAKER_00: in their workflow.
- 13:57SPEAKER_00: And that will take lots of training and lots of time.
- 14:01SPEAKER_01: So I, you know,
- 14:01SPEAKER_01: I have a question in terms of,
- 14:02SPEAKER_01: you know,
- 14:02SPEAKER_01: you guys have been using unit-hipping arrays,
- 14:04SPEAKER_01: just for the listener,
- 14:06SPEAKER_01: for reference,
- 14:06SPEAKER_01: like, you know, these arrays go from, let's say, like, 100,000 to a million.
- 14:10SPEAKER_01: I know some of them go above a million now.
- 14:12SPEAKER_01: There are old-school arrays,
- 14:13SPEAKER_01: old-school as in, like,
- 14:14SPEAKER_01: 15 years that were, like,
- 14:15SPEAKER_01: 100,000,
- 14:16SPEAKER_01: 60,000 that people would use, but,
- 14:18SPEAKER_01: you know,
- 14:18SPEAKER_01: pretty soon they moved to the 650K arrays,
- 14:21SPEAKER_01: you know, all these, and, like, you know, customer arrays, et cetera, et cetera.
- 14:26SPEAKER_01: 500,000 is probably, like, That's a little bit median-ish for the data sets you guys got now.
- 14:31SPEAKER_01: But that's not whole genome sequencing.
- 14:33SPEAKER_01: Whole genome sequencing is 3 billion.
- 14:34SPEAKER_01: 3 billion positions of which about like 4.5 million or so,
- 14:39SPEAKER_01: give or take, are single nucleotide polymorphisms.
- 14:41SPEAKER_01: They're variants.
- 14:42SPEAKER_01: From the reference, from the human reference.
- 14:45SPEAKER_01: And so ideally you would have a whole genome sequence,
- 14:51SPEAKER_01: although the added power and information that you get varies.
- 14:58SPEAKER_01: It varies by the trait and the questions you're asking.
- 15:04SPEAKER_01: And also the sample size.
- 15:05SPEAKER_01: If you're looking for rare variants and you have whole genome sequences
- 15:10SPEAKER_01: and you have 10 samples, well, who cares?
- 15:15SPEAKER_01: You're going to get the exceptional SNPs, but what information is that going to give you, right?
- 15:19SPEAKER_01: Can you talk about the prospects for whole genome sequencing
- 15:21SPEAKER_01: and educational attainment, these sorts of genome economic questions?
- 15:26SPEAKER_00: Yeah, certainly.
- 15:27SPEAKER_00: I mean, that's coming, definitely.
- 15:30SPEAKER_00: I mean, the data is already there with the UK Biobank.
- 15:34SPEAKER_00: And as the cost of sequencing keeps going down,
- 15:37SPEAKER_00: again, my number from a few years ago used to be $300.
- 15:42SPEAKER_00: You probably know better.
- 15:45SPEAKER_00: But that's kind of the rate of improvement on this as the rate
- 15:51SPEAKER_00: of decrease has slowed down, but still.
- 15:53SPEAKER_00: So yes, that's coming.
- 15:56SPEAKER_00: Yeah, for sure.
- 15:57SPEAKER_00: I think in terms of predictive power you get from sequencing, well,
- 15:59SPEAKER_00: that's obviously,
- 16:00SPEAKER_00: that's still an open question,
- 16:02SPEAKER_00: although the best we can go by is like those Grimmel study by Peter Vischer
- 16:07SPEAKER_00: and colleagues where basically they use their GCTA framework or Grimmel with G-R-M-L
- 16:15SPEAKER_00: or something like that,
- 16:16SPEAKER_00: G-R-E-M-L.
- 16:18SPEAKER_00: What they do is basically they look at the genetic relatedness
- 16:21SPEAKER_00: between two random person that are unrelated And see how that relate
- 16:25SPEAKER_00: to their phenotypic relatedness.
- 16:27SPEAKER_00: Let's say their height relatedness, education, attainment relatedness.
- 16:30SPEAKER_00: And so the first time they did this with just SNP data,
- 16:33SPEAKER_00: and they could predict roughly 40% of the height variation.
- 16:37SPEAKER_00: Rather, you know,
- 16:38SPEAKER_00: they would not predict,
- 16:39SPEAKER_00: they could account for some meaning like implied Amount
- 16:42SPEAKER_00: of genetic variation explained by the SNPs was roughly 40%.
- 16:46SPEAKER_00: Then they did this with imputed data.
- 16:48SPEAKER_00: I think they were in the 60%.
- 16:49SPEAKER_00: And then later, they did sequencing data.
- 16:51SPEAKER_00: And they could explain, I think, if I remember, well, 80% variation in height.
- 16:55SPEAKER_00: So that means that the...
- 16:58SPEAKER_00: If you have sequencing data,
- 17:00SPEAKER_00: if you have the whole genome,
- 17:01SPEAKER_00: the whole genome accounts for roughly 80% of variance in height,
- 17:04SPEAKER_00: of course, within ancestry, within European ancestry, and so on.
- 17:07SPEAKER_00: So I don't know that as anyone,
- 17:10SPEAKER_00: and probably someone did,
- 17:11SPEAKER_00: but I'm not aware,
- 17:12SPEAKER_00: I cannot think of it now,
- 17:13SPEAKER_00: that someone has done this with sequencing data for educational attainment.
- 17:17SPEAKER_00: Typically,
- 17:17SPEAKER_00: when you do this for educational attainment,
- 17:19SPEAKER_00: with stuff that I explain,
- 17:20SPEAKER_00: you know, less than 20% of the variation or something around there, just under...
- 17:26SPEAKER_00: And twin studies for educational attainment suggest the irritability
- 17:29SPEAKER_00: of somewhere around 40% plus minus 10.
- 17:31SPEAKER_00: So basically,
- 17:34SPEAKER_00: the sequencing data should allow us to more
- 17:36SPEAKER_00: than double what we can technically predict or explain with SNPs.
- 17:39SPEAKER_00: So it's definitely...
- 17:40SPEAKER_00: We might never be able to fully explain the full irritability of any trait
- 17:43SPEAKER_00: with sequencing data.
- 17:44SPEAKER_00: But I think we'll get, you know,
- 18:10SPEAKER_00: We'll get higher and higher for sure.
- 18:12SPEAKER_00: And that will be an important part of the progress is when we start.
- 18:14SPEAKER_00: Yeah.
- 18:15SPEAKER_01: Right now, wait, wait, did I hear right?
- 18:18SPEAKER_01: Almost 15% of the variance is explained just by genotypic predictors, right?
- 18:25SPEAKER_01: Like, I mean, what does a parasite get up to?
- 18:26SPEAKER_01: Some of these, like, private companies.
- 18:29SPEAKER_00: Oh, you mean for a various-back trait?
- 18:31SPEAKER_00: Yeah, for EDU.
- 18:32SPEAKER_00: Like, for EDU.
- 18:33SPEAKER_00: For EDU, is it like EDU?
- 18:34SPEAKER_00: Yeah, yeah, yeah.
- 18:37SPEAKER_00: For education,
- 18:37SPEAKER_00: I think,
- 18:38SPEAKER_00: you know, I mean, what we have is that, yes, genotypic, like SNP-based predictors.
- 18:44SPEAKER_00: Well, you mean Paul Jenkins index,
- 18:46SPEAKER_00: you mean, or the kind of Grimmel type of percent of variance explained.
- 18:50SPEAKER_00: So Paul Jenkins index for education,
- 18:52SPEAKER_00: And these are genetic predictors that take your SNP data and based on,
- 18:55SPEAKER_00: you know,
- 18:56SPEAKER_00: take the weighted sum of your SNP data,
- 18:58SPEAKER_00: let's say a million SNPs, and the weights are their estimated fact size in GWASs.
- 19:02SPEAKER_00: And then they predict the traits of education.
- 19:04SPEAKER_00: I can roughly explain the latest GWAS.
- 19:07SPEAKER_00: This was 2022 Nature Genetics, EA4, as we call it.
- 19:11SPEAKER_00: We're explaining roughly 14% variation in educational attainment,
- 19:13SPEAKER_00: depending on the predictive data set we're looking at.
- 19:17SPEAKER_00: And the SNP irritability,
- 19:18SPEAKER_00: that's kind of the amount of variation technically that the SNPs can explain, was,
- 19:22SPEAKER_00: I forget the exact number,
- 19:23SPEAKER_00: that's what I was saying earlier,
- 19:24SPEAKER_00: just under 20%, if I'm not mistaken, but around that ballpark, maybe 17, 18.
- 19:30SPEAKER_00: While the twin irritability is roughly 40%,
- 19:32SPEAKER_00: maybe that's an overstate,
- 19:33SPEAKER_00: you know, maybe twin method overstate the irritability, but whatever.
- 19:36SPEAKER_00: So, you know,
- 19:37SPEAKER_00: you can technically double the predictive power by sequencing data,
- 19:41SPEAKER_00: by using the, or at least get something close to that, yeah.
- 19:45SPEAKER_01: And so where are we at?
- 19:46SPEAKER_01: You know,
- 19:47SPEAKER_01: I know that the EDU papers have some stuff about, like,
- 19:50SPEAKER_01: you know,
- 19:50SPEAKER_01: other types of intelligence tests, like,
- 19:51SPEAKER_01: you know,
- 19:52SPEAKER_01: what I'm really curious about is more G-loaded tests,
- 19:54SPEAKER_01: so G is the general intelligence factor, basically
- 19:56SPEAKER_00: IQ, you
- 19:57SPEAKER_01: know,