This podcast is brought to you by the Albany Public Library main branch and the generosity of listeners like you. What is a podcast? God daddy, these people talk as much as you do! Razib Khan's Unsupervised Learning.
Hey, everybody, this is Razib Khan here with the Unsupervised Learning Podcast as usual. And I am here right now with Dr. Alex young Alex, could you introduce yourself?
Hi Razib. I'm Alex Young. I'm a research scientist at the UCLA human genetics department. And I'm also a member of the Social Science Genetic Association Consortium, or SSGAC, which has pioneered the genetic association studies of educational attainment and related traits.
Yeah, you know, I want to get into what you study right now. And, you know, in the social science space, so that's kind of why you're here. But, you know, you said you're the human genetics department, can you give the academic listeners some of your backgrounds so that they know you know, your stuff - to be entirely Frank?
Yeah, so my, my undergrad is in maths and statistics. And I also did computational biology at Cambridge. And I did a detail in genomic medicine and statistics, that's a doctorate in Oxford terms. And I kind of got into the social science world a little bit by accident. I lived in Iceland for a while, where I developed methods for estimating heritability. And it turned out that those methods had the most interesting application in education and behavioral traits. And I've ended up working with economists and social sciences scientists and getting into this whole intersection between genetic and social sciences.
Yeah, yeah. So you know, I just want to make clear to the listener, you have a background in quantitative fields, you have a background in conventional human genetics, biomedical genetics, the sort of stuff that is respectable, and now you have moved into this, you know, field that "don't get no respect" from a lot of people from what I can see online. And so that's what I want to talk to you. What is your current project? What are you currently working on so that listeners can get a sense?
So we've been working on an update to the genetic association study of educational attainment. So this is going to be one of the biggest genetic studies of any trait, it's going to have 3 million people in it. I'm also working on methods. So I developed statistical methods and software that try and disentangle nature and nurture using family data. And we're also putting together a lot of genetic family data to try and answer some of these trickier questions about nature and nurture and their relation.
Yeah, let's, let's get a few definitions on the table for the listener. Heritability. Can you tell them in words that a normal person on the street to kind of understand what heritability is in the narrow sense and in the broad sense, and then I will stipulate, when we say heritability From now on, I think we're generally going to be talking in the narrow sense but narrow sense, but I just want to clarify that.
Yes. So heritability tries to measure the contribution of genetic differences to the variation we see in human traits. So we observed that people vary in their heights. And one question we can ask is, what fraction of that variation in heights in the population is viewed to genetically cause differences? And there's a distinction between narrow and broad heritability. So what I said before is really broad heritability is any kind of genetic effect. So it's measuring the total contribution of genetic effects to variation in the population for a particular phenotype, like height or education. But typically, we study narrow heritability, which restricts the kind of genetic effects we look at to ones that we can compute just by adding up the effects of individual genetic variations spread across the genome, like a linear model, essentially, whereas broad heritability would include interactions between different genetic variants. There's some controversy about whether there's much difference between the broad and narrow heritability but I think there's a lot of evidence now that basically the narrow version captures almost everything.
Okay, so you mentioned the height, and then you mentioned education. And a lot of people would say, I mean, height is a phenotype If that's easy to define, you just take a ruler or tape measure, you measure them, you look at this distribution within the population, then you look at the variation in genetics. And we'll get into, you know, classical designs using twins and correlation of relatedness. Going back to Fisher, R.A Fisher, and what we do now what you do now with genomics, but and then with educational attainment, okay, like, what is this education is socially constructed? You have the number of years, how do you compare across institutions, you know, countries, all these other things, I know your latest work is going to have a sample size of 3.1 million. Obviously, that's going to vary a lot in terms of what, you know, three years of education beyond secondary school mean, you know, so can you unpack a little bit of the skepticism that's coming at you from that direction?
Yeah, so obviously, height is much easier to measure trait than educational attainment, and it's not so contingent on the particular culture within which it is measured. So in some sense, we don't necessarily study this trait, educational attainment, which is defined by the number of years of schooling that you obtain. So whether you complete high school, go to college, do a PhD, we don't necessarily study that trait, because it's the best trait to study, we study that trait, because it's easy to get really large sample sizes for. So there's always going to be some complexity and trying to make this definition equivalent across different countries. And we use these ISCD educational categories to try and make equivalences to US years of schooling across different cultures. But that's always going to be somewhat perfect, imperfect, and there's going to be heterogeneity. But by getting these really large sample sizes, we can maximize our power and build predictors that have the maximum amount of predictive ability.
So yeah, let me let me just rephrase some of what you said, or like, put my interpretive spin. And you can correct me doctor if I if I get this wrong. Basically, I think one thing that I think about it, the way I think about it is some of these imprecise characteristics where the phenotype is not as clear and distinct. In a way, when you're calculating heritability off these, it's actually a low, it's kind of a floor of what the heritability of the underlying bundle of characteristics could be. So if you use Raven's Progressive Matrices, which is a highly G loaded intelligence test, you'd get really, really precise estimate of what you're looking at there. Whereas with education, there are cases. You know, I mean, look, to be frank, there's cases of social promotion, there's cases of fraud, there's all sorts of issues related to education, that are totally unrelated to one's individual underlying genetic characteristics or aptitude. Sometimes there is, you know, a systematic bias within a certain region, due to a policy change. So all of these would introduce environmental noise and stochasticity into your models and reduce the heritability I mean, that that's my intuition. What do you say to that?
Yeah, so educational attainment is a very downstream trait is very downstream of the underlying biology. So it's going to be less heritable than the traits are upstream of it, such as cognitive ability, and probably other personality related traits that are important for going for an educational attainment. And as you were saying, we know that there is an environmental component to educational attainment that's supported by old school behavior, genetic studies, they find that, you know, that family environment that siblings share that does contribute to educational attainment. Whereas if you look at something like cognitive ability, or executive function, those those kinds of traits, there's a lot less evidence that the family environment actually affects us.
Yeah, let me ask you something about executive function. In fact, let's talk about heritability is of various characteristics. In Dr. Paige Harden's, Katherine Page Harden's book, the genetic lottery, she was reporting. You know, I don't know this literature, obviously, as well as you I don't know as well at all at this point. She was reporting heritability of executive function of almost one almost 100%. I mean, what's up with that?
I also found that surprising. I'm not sure if, if that holds in adulthood, I think that was looking at kids. But in general, we see that heritability increases for these cognitive and behavioral traits into adulthood. So I'm not sure if that study is being done in adults as well. But I guess this there's pretty good evidence that the executive function is among the most heritable of cognitive abilities.
Okay. All right. That's That's good to know. And so let's let's talk about let's talk about what the heritability of various characteristics are. So you know, You know, we've we've actually done a podcast, my previous podcast, the insight, I'll probably put it in the show notes. We you know, we talked about heritability and nurture, genetic nurture these sorts of issues. And so you know, heritability of height, the estimates go from what, like 60% to 90%. Heritability of intelligence, probably like 30 to 70 or 80%, depending on who you listen to these sorts of things. Can you just rattle off some heritability statistics for, say, psychological phenotypes? Is that a few physical characteristics?
Yeah, so I think personality traits are usually around 50% educational attainment in terms of years of education, depending on who you listen to. It's between 20 and 40%. But there is a role of the family environment, cognitive abilities are probably in the range of 50 to 60%. I guess this executive function trait is even more heritable. Then, you know, traits, people might be somewhat surprised that are quite heritable, such as the age at which you have your first child, as a woman that's between 20 and 30%, heritable. I mean, some physiological and appearance related traits are close to 100%, heritable, like eye color, and fingerprints. And I guess we should also say some traits are not heritable at all, like what accent you have that is transmitted from parent to offspring, or at least some correlation between.
Your parents don't sound like you.
Yeah, no, I'm saying that they do there is this correlation between often there is this correlation between parent and offspring for accent, but it's, it's a function of cultural transmission, rather than genetic transmission. And I guess something like educational attainment that has both genetic transmission and cultural transmission, some things such as cultural transmission, like what particular religion you're in, some things are completely genetic, like your eye color.
Yeah. And just to review for the, for the listener, you know, heritability, one way you can think of it is what's the proportion of the variation in the phenotypic, you know, distribution of the population that can be explained by variation in genes. And so heritability of like, 100% is like, okay, the variation of genes is explaining all the variation. That doesn't mean we have some sort of deterministic, you know, system here, there's still variation, you know, with recombination and other things and pedigrees. And another issue is, just want to put it out there that, you know, from an evolutionary genetic perspective, often highly heritable traits actually have a low fitness implication, because they're variable of the population, which means that the variation hasn't been squeezed out, creates traits with extremely low heritability are often extremely important for Reproductive fitness or fitness in general. And so those characteristics like there's not that much variation left, because there's negative selection against the variable types that aren't as fit. So these are just some intuitions I want to put out there for the listener, as we kind of dig into the weeds here. Next, I want to I want to ask you about, you know, Polygenic, traits, genetic architecture, like the distribution of these traits. So obviously, I think most listeners know what Mendelion traits are traits, like at single gene, say, with cystic fibrosis, you get two copies of the gene, one from your mom, one from your dad, you know, you got some issues with your lungs, you know, similarly, with sickle cell anemia, you know, you get two copies from your parents. And there's a problem. Okay, those are Mendelian genes. But we've been talking about here, and what we will probably talk about as we go forward, are polygenetic characteristics, these little effect, genetic, you know, effects plus and minus across the genome. And can you just give us a quick survey of like, what's been going on over the last 20 years and our understanding of the pervasiveness, these characteristics of the genetic architecture of the effect size distributions within the genome for various traits, like height, intelligence, educational attainment, and, you know, like what we've discovered?
Yeah, so I think that there is a bit of an issue in the public understanding of these things. And I remember my high school biology, I think all that I was taught about was cystic fibrosis and how that was inherited. And I think that's kind of the model that most people have, in their mind that there's, there's a gene for something, just one gene. And that's not the case. For Complex traits, such as height, and education, there are probably 1000s of genes, or at least definitely 1000s of genetic variants that affect these traits, each one of which has small effects. And this, this has made it a lot harder to figure out which particular genes are actually affecting these kinds of traits. So I mean, just to give an example, for body mass index, the common genetic variation that explains the most BMI variation in the population, only explains around half a percent of the BMI variation in the population, and that's actually considered quite exceptional, exceptional in complex human traits. And, and there's nothing like that in educational attainment, for example, there's nothing that explains so much variation. And when we've done these association studies of educational attainment, the typical kind of variant that we discover affecting educational attainment, having the education increasing version of that variant is only going to increase how much schooling you get by about two weeks on average. So these are really tiny effects that are spread out across the genome, probably 10s, of 1000s of variations, each one of which is not very predictive of how much education you have. So you can't simply look at someone's genome and see, they have this one particular variant, therefore, they're likely to be really tall, or to get more education, you have to look across the whole genome and add up of all of these real small effects to understand the genetic contribution to trait.
Yeah, and so for the listener, you know, all of what, what Alex was talking about, you know, was theoretically understood, it was theoretically understood 150 years ago, or, you know, over a century ago, okay, with biometrics, biometrical analysis, you know, the application of statistics to evolutionary and biological questions, and eventually fusion with Mendelanism. But experimentally, this was not really you, it was not tractable people were using. So for example, Drosophila studies, use recombination recombination units associated with physical markers in these breeding programs, to figure out genetic locations and associations very tedious. You know, more recently, you know, people have been using these pedigrees, you know, the probands, and whatnot, you know, the stuff you see in medical genetics, but those sample sizes are small, they're gonna pick up the big effect sizes of Mendelian stuff, there's just really no way to explore quantitative traits, until we had genomic. So I want to get Alex on because he is a genomics person. So yes, he does do social sciences, he does do behavioral genetics, but he is a genomics person. So genomics is there are 3 billion base pairs in your genome of those depending on how you want to count polymorphism, you have on the order of 10 to 15 million polymorphisms. Of those closer to around 10 million, maybe five to 10 million are common within the population. So only a tiny proportion of your genome is variable. And many of those tiny proportion, many of those variable genes are alleles. genetic variants are found in other individuals. And so they're looking for these common variants that are explaining the variation within the population. In some cases, there are probably rare variants that are also explaining the variation within the population. But all of this is only possible with modern technology, with sequencing machines, you know, that are popping out these strands, these reads, and then you have, you know, massive, you know, computational power to reassemble the strands and create this digital sequence. And that's only been possible over the last 20 years, it's really only been feasible very much over the last 10 years. And all this social science stuff that Alex is talking about the genetic architecture is so diffuse, it really wasn't feasible much before 2015. So all of this has happened to the last five years or so. I mean, do you think that's a correct summation? Alex?
Yeah, Yeah, I do. And I think that people actually, maybe they shouldn't have had this expectation. But people had this expectation that the architecture was going to be simpler, and we just have to get a few 1000 people with that genetic data and their traits, and we'd figure out the genes that affect all of these complex human traits and diseases. But based on population genetic theory, if there is selection, constraining genetic variation, so genetic variants that have a large effect on on getting a disease, especially a disease that hits you early in childhood, those are going to be selected against and those frequencies are going to be pushed down. And that means these these kinds of genetic variants that maybe directly alter a protein, and thereby have a really large impact on on your phenotype, those tend to be low frequencies. And that means it's, it's hard to discover these these big effects on on human traits. And typically, with the kind of data that we have, currently, and especially for large samples, we're only able to interrogate these common genetic variations that typically have small effects on complex traits, like height and education.
Mmhmm, yeah. So um, you know, speaking of small effects, you know, I do want to ask you 2017, Boil et al. I think the Pritchard Lab came out with this omnigenetic model, which is basically starting to converge on Ronald Fisher's infinitesimal model where there's like, you know, these innumerable numerable small effects and there's some core biologically relevant genes. But does that does any of that affect your model?
So I think it depends a bit on the goal of your study. So if your goal is to discover biological mechanisms, then yes, I think that should affect the way you conduct and design your studies. If your goal is to predict phenotypes, then you're less worried about what the actual underlying biological mechanisms are. And for a lot of the work we've been doing in the Social Science Genic Association Consortium, it's about constructing genetic predictors of traits that can be used in social science studies. And for that, the key issue is to get the biggest possible sample size. And we want to build these predictive models. And we're and we can be somewhat agnostic about the biology linking the genetic variation to the ultimate outcome. However, especially in medical genetics, and disease, genetics, it's a lot more important to actually get some kind of genetic understanding this, this particular protein is translated into into some phenotypes through some some intermediate interactions in the cell, or whatever. And maybe that can be a drug target. And I think that for that kind of work, where the field is going now is using genetic data that can identify protein altering variants, which tend to be rarer in the population. So this is whole genome sequence data and whole exome data, which is more expensive to generate. But a lot of pharma companies are betting big on this as a way to understand biological mechanisms of disease and design drugs to tackle them.
Yeah, let me let me let me back up for the for the listener real quick. The whole genome, obviously, you know, you guys know what that is the whole genome, usually, to get a medical grade whole genome, you do something called 30x coverage. So it's really high quality readout, where you sample every position 30 times so you don't get false positives, okay? Well, Alex just said exome exome is like about the 1% or so of the genome on the order of 1%, that actually codes for proteins. So that's like the functional genome - now there are regulatory units. There's other things going on structurally, outside of that - the intergenic regions, but really, a lot of the core action happens in the exome. And so that's why he mentioned that. So I guess, you know, a question I want to ask a couple of years ago, three years ago, now, you know, getting old, you know, Peter Visscher, and his group and Peter Visscher is in Australia, they do a lot of work overlapping with your work, whole genome, quantitative traits, you know, kind of fusing quantitative genetics, the field of quantitative genetics, which kind of applies statistics to genetics, really, you know, plant breeding these sorts of things along with genomics. And they kind of argued that the missing heritability problem, which is, you know, the variance, you know, so we know from classical studies, and I do want to get into that, that the variation, due to genetic variation of given traits could be 50%, as of the earlier genetic work, was only capturing 10%, which meant, like, Where's the other 40% of the variation? You know, some people were claiming, oh, that was just a spurious It was artifactual. This is the real amount of variation or other people are saying, okay, we get like, more resources put into it move from like chips that only cover a small proportion of the genome and tag only a small number, you know, they should tag most of the genome, depending on the LD, I don't want to get into that I think the listeners will get bored. But um, you know, basically, do whole genome sequence have large samples, and you'll get the missing heritability. What do you think about that argument? Because I mean, I guess that's predicated on the idea that most of the heritability in these traits of interest is kind of like an additive narrow sense heritability.
Yeah, so I kind of agree that probably most of the heritability is an additive narrow sense heritability. I think there's still an open question as to whether twin estimates of heritability are too high. So twin, traditional twin studies, they compare the similarity of identical twins to non identical twins. And the assumption is there any greater similarity of identical versus non identical twins, is due to genetic effects. And that assumption is very hard to test. And some people think that it's it's likely to be false, and that twin estimates are too high. I developed a method for estimating heritability in Iceland that doesn't make so many assumptions. And the estimates I got were lower than what you get out of twin estimates. There could be other reasons for that. There are some other ways of estimating heritability from genomic data that I think also indicate that perhaps the twin estimates are a little too high. But it's, it's still an open question. In my opinion, I think that the whole genome sequence data is going to help us determine that. But I think that we still are lacking the the really large sample sizes that we need and ideally also family data. Because if you don't have family data, it's hard to distinguish what's the real genetic cause from some kind of confounding? And I think that is that is an issue that is hard to solve with based on current data.
Well, why don't you? Why don't you talk about a little bit just like, you know, tell the listener what we mean by twin design, how that, you know, monozygotic / dizygotic , this sort of stuff, how that works, these classic old school family design, and how you guys are doing it now. And then let's talk about, you know, the family studies that people want to do. by tracking, I think what you're talking about implicitly, is a genetic variation among siblings even helping trying to figure out some of these compounds.
Yes, so the classical twin design is comparing the correlation of identical twins, which are genetically identical to the correlation of dizygotic twins, which are related as normal siblings are related. So they're 50% identical. And by comparing the phenotypic correlations that have phenotypically similar, there's two different classes of twins, you can estimate heritability. And that's generally been seen as the gold standard way of estimating heritability. However, it does make this assumption that there isn't something unusual going on with identical twins versus non identical twins, it just makes them more similar. Maybe they, they interact with each other in in this quite profound way, because they're identical twins, it's not shared by non identical twins, and that that can cause some spurious inflation of the heritability estimation. So Peter V isscher, actually, back in 2006, he developed this method that takes advantage of the random variation of genetics within a family. And that's something that I've been exploiting a lot in my work as well. So whether you inherit one or another copy of a particular genetic variant from your mother, and from your father, it's like the outcome of a coin toss, it's just this random thing that's uncorrelated with the environment. So if you can exploit that random genetic variation within a family, you can get estimates of heritability and genetic effects, there are unbiased that they're removed from all the confines, it's like running a randomized control trial in medicine, you're like randomly assigning one sibling to have one genotype and another sibling to have the other genotype. And those siblings share their parents, their household, the neighborhood, they're in often the school they go to. So it's, it's a very well controlled design. If you think about just general design of scientific experiments. It's a natural experiment. But it's a very well controlled natural experiment. And that's the variation that I've tried to exploit in my work. And at least, I think, based on my work, and some other work, that there is probably some inflation of heritability in twin estimates. But it's not that the twin estimates are completely wrong, it's more that maybe they're just a little bit too high. It's possible that there's some kind of esoteric genetic variation that is not being captured by the genomic technologies at the minute that explains this gap, such as genetic interactions, structural, complex structural variations in the genomes, such as repeats, or copy number variations. So it's, it's still something of an open question, what, what, what's going on there, but I think that the future will be looking at family data, genetic family data, because before in behavior, genetics, we didn't have genetic data, there's only so much you can learn about genetics without genetic data. Now we can integrate some of the ideas that go back to twin designs with modern genetic data. And that allows us to really discriminate things with a lot more power.
Well, let me let me make it a little concrete for the listener too - what you're getting out here. So you know, I think in 19 , 18, paper by R.A . Fisher, the supposition of you know, a correlation between relatives, whatever, it shows that Okay, so relatives, your aunts half your genes with your parents 1/4, with your grandparents 1/4, with your aunts and uncle's, an eighth with your first cousins, half with your siblings. So you see these patterns of relatedness patterns are variants, these patterns of relatedness and variance should be correlated with genetic variation and direct proportional to the relatedness. So these are assumptions you make with a model. The twin study is kind of like a specific case of this, I think, and you know, you have these assumptions. They're pretty robust on average, and then you use that to figure out what's going on. What you're getting out now - is today, okay, you are about 50, you are 50% related to your parents like that's determined that's deterministic as per, you know, Mendel's law. assuming there's no segregation, distortion, you expect 25% to each grandparent, but there's actually variation in this. So for example, I can tell you empirically, that my kids vary from 30% to 20%. On any given, you know, comparison with grandparents, I have known sibling pairs that are 40% related, and sibling pairs that are 50% 58% related. So there's a fair amount of variation around that 50%. And so now, you know, as a geneticist, you know, exactly the proportion across two siblings, instead of just defining them as full siblings, you can call them, point five for siblings, or .44 siblings. And so you get a much better quantitative tracking, well, not much better, but on the margin, a considerably better quantitative tracking of the genomic relatedness with a phenotype. And that's the power that you're getting. I do want to ask you, a couple years ago, Molly Shorski's lab came out with a paper about heritability estimates. And how, you know, there's some sort of ecological confound, even in the British biobank when you control for population stratification. And by population stratification, I'm talking about race population structure. For the listener, basically, you want a genetically homogenous population, ideally, to get the signal out of there. Because otherwise there'll be all these other variations that will be correlated with population ancestry. But even in the white British biobank sample, which is genetically quite homogenous. Molly Shorskis group found that the heritability might not be accounting for all of the compounds like what do you say to that? I did ask, I think Amit Kira, and he was actually excited. He's worked in this in the cardiovascular field, he was just excited, cuz he's just, I just want to, I want to control all the environmental stuff, too. So why wouldn't I want to know this? Right? So what do you think about that?
Yeah, so I agree with with Molly on that, we actually wrote a review paper for Science together along with with Augie Calm my my postdoc supervisor, and I think we share a perspective on this. And the way that I think about it is that the fundamental well controlled group within group comparison is within the family, because within the family genotypes are randomly assigned. So that's when you can actually make something approaching a causal inference about genetics, that this particular genetic variant you inherit really causes this trait to be different. Once you expand out from the family, things can become confounded. So if you go out from within a particular sib-ship, essentially, if you build people that are siblings, you go to the wider family, even there, things can be confounded, because your environment is related to your genetics. And that means that the, you're doing this between group comparison essentially, where the groups are genetically different, but they're also environmentally different. And as you expand that circle further and further out, you're going to get more of this contact in between genetics and environment. And at the level of the UK Biobank sort of white British population, that's still fairly homogenous and global terms. But there is clusters, genetic clusters in different regions of the UK, the UK has a social class structure. And all of this is going to be correlated with genetics. So you have this complex interweaving of genetics and environment at the level of a country or the level of the globe. And that can confound your genetic studies. But if you can do stuff within a family, you don't have to worry about all that confounding. And so I think that, that use of the the genetic variation within a family which is random, that's what really enables you to distinguish causality from content in genetic studies.
Man, you talk about families and the values of families more than the Christian coalition in the 1990s. It's all good. I think you're, you know, I share your perspective. When I saw the Visscher paper, I'll link to the show notes, you know, in 2006. And I know some people mutual friends of ours are but to be frank, not big fans of that paper. They think it was a bit sloppy, but the underlying logic, I hope Peters not listening, but the underlying logic. It seems pretty straightforward and strong. In terms of Okay, family based studies. Do you think there are still some serious confounds, though with say, like EDU or something like that within family based studies, like systematic biases within families? Like I don't know, maybe they treat the better looking kids better? I don't know. There's still stuff like that that needs to be accounted for correct?
Yes. So this kind of goes To a thought experiment by Yanks who is a Harvard sociologist in the 1970s, he proposed this thought experiment about a gene for red haired people. In a society where there was discrimination against people with red hair, they were denied educational opportunities. And the Yanks thought experiment was useful in the the way that behavior genetics models are set up, that would actually count as part of the heritability. In such a society where there was discrimination against people with red hair, you inherit the Red Haired gene, in some sense, it does cause you to get less education, but it only causes you to get less education through a social mechanism of discrimination that is perhaps not biologically meaningful. So some people have taken that thought experiment and used to try and write off essentially the entire field of behavioral genetics, or to ascribe the entirety of the heritability to mechanisms such as discrimination based on appearance. And I think that kind of argument is really not tenable based on our current state of knowledge. While maybe when you only have social science, data, correlations between relatives, it's harder to get at these mechanistic things, it's still hard to get a mechanistic things with the genetic association studies that we've done so far, but we have actually learned some things that allow us to reject that kind of mechanism as a primary mechanism for the heritability of educational attainment. So in the last published, genetic sociation study of education, what they're able to do is identify genes that are close by to the particular genetic variants, that effect that we found are associated with educational attainment. And they can see in which tissues, those genes tend to be expressed in relative to some background level of expression for random set of genes. So we're able to identify particular tissues that are driving this genetic association with educational attainment. And if you do that, you see the nervous system lights up, that's that's the main signal is the nervous system. If If this mechanism was going through physical appearance based discrimination, it would show up something to do with things expressed in the skin and the hair, but instead, it shows the nervous system. And it identifies genes involved in in neurons as well actually not in other cells in the nervous system, glial cells, which are building these structures around the neuron. So we have learned something about the likely broadscale kind of tissues and pathways that these genetic effects acting through. And it really doesn't look like they're acting through these mechanisms of discrimination. So I think that kind of argument just doesn't really hold up against the modern genetic data and knowledge that we've gained from it.
Yeah, so let me unpack and explore this a little bit. Um, so a common opinion or a common assertion that I do get from evolutionary geneticists to be concrete, for example, is that Well, I mean, you know, maybe they're just like picking up skin color, but we actually know the genetic architecture of skin color pretty well, unlike, unlike height, or even more, so intelligence its polygenic. But it's really dominated by a few alleles of relatively large effect. You know, so often they're associated with epithelial development cells, you know, like cells associated with a membrane transport, because that's a correlate with melanasome melanosomes production and development. So Soc. 2045, is a huge hit in most world populations that have variability in skin color. It explains 30 to 40% of the skin color difference between African Americans and Europeans when you just do that comparison, SLC 45 another solid carrier is the next biggest one. There's others like OCA2 HERC2 to express eye color and skin color for red hair. There's a bunch of necessary conditions but often the variants can be explained for my melana carnton one which is found in other mammals as well. So we know these alphabet soup of genes and to my knowledge none of these have shown up in hits in educational attainment Am I wrong?
No. They haven't shown up and we would know by now if that if that Yeah, because those things are pretty strong effect. Variants way stronger than anything we see on educational attainment that's not what shows uh, what shows up is is snips is genetic variants. The To close to genes that tend to be expressed in in nervous tissue, which is kind of makes sense, really. I mean, if you do the same analysis for hight, you see muscular skeletal tissue. If you do it for educational attainment, you see the nervous tissue show up because education is probably related to what goes on in the brain. Yeah, seems kinda obvious to me.
Well, let me let me let me be more concrete. You know, I did a podcast and I'll link to this one too, with a mutual friend Daniel Crouch, who has worked with Walter Bodmer on facial morphology. Now facial morphology is not quite as simple. As, as you know, skin color skin color, like let's say about like 20. In worldwide pool samples and stylized fact, it's about true that 20% or 50% of the variance can be accounted for 20 quantitative trait loci, right, the other 50% is, you know, distributed on smaller effect. morphologies, a little bit more diffused, but it's not nearly as diffuse as height. And so again, they tend to be enriched for bone morphogenesis genes. That make sense, right? Our bone morphogenesis genes are is, you know, I basically asking, does visible, non European physical ancestry on your face? Is that correlating with any hits that you see an Edu?
Well, no. And we also wouldn't, you also have to take into account what sample we do these studies in. Because we restrict the samples generally to people with vast, predominantly European genetic ancestry. And actually, most of the work that I did on on heritability, and was done in Iceland, and we restricted that sample to people of European genetic ancestry that have all four grandparents born in Iceland, and Iceland is a pretty homogenous, egalitarian country. It has one of the lowest levels of income inequality in the developed world. As a Brit walking around like Reykjavik, it's kind of like very different walking around London or out LA. Now you do
Wait, are you trying to say that Iceland has no systemic racism? Is that what you're trying to say?
Well, there's there's not so much in the way of racial variation is especially within people who have European genetic ancestry and all four grandparents born in Iceland, I mean, almost everyone is is is northern looks Northern European there, obviously, this variation and in an eye color and hair color, but I doubt that there's this really strong, dicrimination On the basis of these things in such a gala, terian minded, community minded kind of country.
Yeah. Well, so I mean, you know, I like what's what's what's lurking in the background, we haven't explicitly said Iceland. They are descended from common founders. It's not a migration country traditionally. And so all Icelanders have their genealogies and they've been genotyped, by decode company you work for or you work with. And so you know, they're all about third cousins, or fourth cousins, or fifth cousins, something like that, right? So this is as homogenous as you can get as a sample, I want to be concrete. And I'm being specific, because these objections keep coming up by people who clearly have no idea what the data sets are, what the data results are. But they're quite often blown up on social media as awesome dunks, and it's extremely frustrating for me, I assume that it's much more frustrating for you, because this is your work. I want to hit these questions, even though they, frankly sound dumb to me, because they come up over and over again with people who are PhDs in population and evolutionary genetics. Okay, so for the listener and Alex, that's why I'm asking these bizarre questions.
Yeah, look, I mean, if you're doing a within family study in Iceland, restricted to people with all four grandparents born in Iceland, these factors are not important. I do you think that in the genome, the association studies that we've done in educational attainment in more diverse populations, but that they're still in people of European genetic ancestry, but they're more diverse in Iceland, and in the UK? Sure. There's, there's social stratification, there's regional inequalities, and I think they do generate some bias in the standard genetic association study design. But when you're doing within family study and a marginalized population, these things don't apply. You know, it's it's this mechanism is not it's not relevant.
Well, okay. Well, let's, um, let me let me get to another question. Again, that's not my question, but kind of stuff that comes up. You guys cannot do things like common garden experiments, like you can't experiment on human beings. And so you can't discuss, you know, the argument is you can't just distinguish sociological explanations from biological ones because you can't do a controlled experiment. Like you just have these correlations, like Well, what do you say to that?
So it's true The mechanism through which a particular genetic variant leads to one person getting more educational attainment and someone else with a different version of the genetic variant, that can be very complex. And it can be socially mediated in some sense. I mean, genes on their own, don't do anything. If you haven't have a bunch of chromosomes in a petri dish, it's it's not going to solve any differential equations, right? I mean, the fact that genetics exerts its effects through the society in which we live is not a reason to deny that genetic variation between people is important for explaining differences in how well they do at school. And I think that, yeah, these things are more complex than a simple monogenic Mendelian disease or something. But that's not to say that there aren't real genetic causes of differences between people. And they might be socially contingent to some degree, it's worth pointing out that heritability is not a, it's not a fixed parameter. If you change the school system, the heritability might change, the particular genetic variants that lead to success or failure might change. But there's still going to be genetic variation there that the effects how people do better or worse. And in school, it's just it might be slightly different, it might have a bigger effect or a smaller effect in total, but it's still going to be there.
What Yeah, and just kind of like stepping out of this persona for a second, the stringency that they are asking, is never asked for any other type of social science that they're using to refute your own results. So you know what I'm saying like me, you actually have like more stringent criteria that a lot of sociological results and yet, you know, they're rejecting your results, and they're saying the sociological results should be preferred. And so that's kind of a peculiar thing. What about the idea that the field produces zero biological insight?
Well, I think that it would be great if we had more biological insight then we do. I don't think it's a criticism that should only be levied genetic studies of behavioral traits. I think it's also a criticism that has been levied at genetic studies of physiological and medical traits, there was a lot of hope that we would really figure everything out really quickly with doing these genome wide association studies and in relatively modest sample sizes. And it hasn't turned out to be that way, in part because of this extreme polygenicity, that we have all these variants across the genome, and it's hard to figure out what they're doing. But that doesn't mean it's produced no biological insights. As I was saying before, we see that genes expressed in nervous tissue light up as being important. I mean, I guess one can argue that's not a particularly groundbreaking insight that the nervous tissue is important in how people do educationally. I think one insight that came out of the 2018 education paper was the genes expressed in neurons appear to be important, but not in glial cells. And glial cells are important in determining things like neural processing speed. So, you know, there's some kind of insight coming from it, but it's perhaps not as much as we would like. But just because biology is complex, doesn't mean that it's not there, it's not operating in some way. And this field is still really in its infancy, I think that it's going to take longer and be harder to figure out the underlying biology than then we might have liked. But I have optimism that with, you know, these new kinds of data becoming available that the exome, the interrogating that in detail, the protein coding part of the genome, for really large samples, that's going to make a big difference for this, we're going to be able to much more easily figure out some kind of story that goes from altering a protein to some downstream behavioral trait. And actually beyond education, that are behavioral traits, like addiction, where we have insights from these studies, there's there's a common variation in the nicotine receptor gene that's important in determining how much people smoke and addiction is obviously something that's mediated through complex social pathways, but the fact that you have a different version of the nicotine receptor is also important. And similarly, if you have a broken version of the alcohol dehydrogenase gene that quite a lot of people at East Asian ancestry do that causes the flushing reaction, you're in low risk of alcoholism. alcoholism is obviously much more complex and one particular gene but doesn't mean that genes on important to?
Yeah, I mean, I think that's, I mean, look, I agree with you. I'm just, I'm just making you answer the questions for the people out there, you know, so that they know what's going on. And you know, just to be explicit and clear, in case people don't understand what you're talking about new resources, you guys have been working mostly with Gene chips with snip array chips, that have 500,000 to a million markers, depending on the chip. And these markers are distributed, you know, across the genome, they're enriched for certain functional positions, but they're also distributed across the genome. So you can tag and capture variants along the genome that are known to be associated with the market, correct? Yep. Yeah. So some of the issues that we're having that, you know, Alex and his colleagues are having is due to the fact that they're not having a full representation of genetic variation, they're having a subset. This was a technology that was state of the art in 2010. And it was extremely expensive. And it was great that they can have access to it in 2015. But the way that the field has changed means that now they can go to whole genome sequences proceeding into the 2020s. And perhaps they will find nothing of biological significance, like their skeptics say, but we will know we will find out right?
Yeah, we'll find out. And pretty soon to be honest, the UK Biobank, which has a sample size of 500,000, they're going to have the whole exome data for that whole sample. So we're going to be able to see what proteins are important for all these complex traits. So we're going to have much better power than we ever had before to do this, and they're already exciting results coming out from that, that study more in the in the medical domain at the minute, but there's no reason why those methods can't be applied behavioral traits as well.
Alright, so I mean, you, you know, you've done some work, you kind of, you know, made your name, I guess, like probably your your biggest, I think individual contribution, maybe at least to the public, is the genetic nurture stuff. Can you tell us what genetic nurture is so that people know, and how this fits into the broader picture of EDU and the other things you're studying?
Yes, so genetic nurture, was a concept that that primarily was developed by by my postdoc supervisor, Augie Kong, although similar concepts have existed in evolutionary genetics. And then plant and animal breeding were not really plant to suppose but more animal breeding, where you can have an effect of the genetics of the mother or the father on the trade of the offspring through the environment provided for you by by the parent. So in the context of humans, you could imagine having a genetic variant that raises your educational attainment on average, and you inherit this from one of your parents. So if it raise your educational attainment had some effect on on you, your cognition or something like that, and that affected the educational attainment, it likely also had that effect on at least one of your parents. And if your parents have more education, maybe they were wealthier, they valued education more in that environment they provided for you is going to affect your educational attainment too. So a genetic variant that you inherit can have what we call a direct effect on you from actually inheriting it and having some biochemical impact on your development. They can also have an environmental effect on you through the environment provided for you by your parents. And this was something we investigated using the Icelandic data, which is quite unique and having large numbers of individuals with genetic data on their parents. So we could look not only at the genes that were transmitted to the offspring, but also the ones that weren't. And we could correlate the genes that weren't transmitted to the offspring with the offsprings educational attainment. And we could see that even the genes not transmitted to the offspring, were predictive of the offsprings educational attainment. And this surprised the field, I think, although perhaps it shouldn't have. And it showed that, you know, around only around one half of the, of the variance explained by genetic predictors of educational attainment was due to the direct effect. And some other fraction of it is probably due to this generic nurture phenomenon, that the genes of your family, your relatives are affecting you through the environment.
Yeah, I mean, this is subtle, but it's important. And you know, as you know, as you note, you know, the broader concept, had actually been around for a long time. So one of the things that I want to mention to the listener is that, you know, a lot of these theories, a lot of these concepts, a lot of these frameworks are old, they've been around since the early 20th century, it's just a miracle. And I'm gonna use the word miracle, the M word of genomic technology has really allowed to decompose and explore it right now, in the modern, you know, applied sense as opposed to just just a theoretical framework. So I think like stuff like that has changed a lot. Um, I want to ask you like, I want to You know, a little bit of controversial question. Um, so you know, Dr. Katherine Paige hardens books come out came out recently. And um, you know, it's made a big splash I, you know, listeners probably know of a review and unheard, I also have a much longer review on the substack, you should check it out. And so Molly Shoreski, who you've co authored with, she had a Twitter thread about heritability in between group heritability. And, you know, basically, pretty gently, I thought criticized, Dr. hardens. argument that you cannot compare heritability. Not necessarily between groups, but the heritability of traits in one group, you can't just predict, predict, project them into the other group, because of, you know, issues. The portability of GE was, which I've talked to with Alicia Martin, and to some extent you I think, in previous podcasts, and all of these things. But you know, she was basically I think, getting at the fact that, you know, the idea of populations are themselves an artificial construct, and, you know, kind of like you were saying anything beyond the family, it's, you got to be careful there. And, you know, I think, I know, I think Paige was a little taken aback by that, because geneticists, population geneticists have been saying something like this as a mantra for decades. And this, this flip was a little, I think, surprising to her because she's comes out of a psychology background, and she just taking what people say, right? They've been saying this for decades. What is your take on that? And I will say, I retweeted it. But people who I don't agree with at all, who probably think the heritability of EDU is zero, also retweeted it. So this was a tweet thread where it kind of provided everything to all comers. You know, and so I mean, what did it provide to you?
Yeah, I agree with you. I thought it was quite interesting that people who like to minimize heritability, really liked that, but I'm not sure. That was really Molly's point, to be honest. I mean, she should speak for herself, really, I don't want to I don't want to put words in her mouth. But But the way that I read it was, is more point that there's some kind of scientific incoherence in saying that this is a group, and we can do stuff within this group, and it's fine. But with that, within that group, you can define all these subgroups. So like, what is what is the population really like? Yeah, population geneticists, they have these models, and you have like, one population and other population, like they're on different islands. But in the real world, it's it's not like that, right? I mean, there's, there's no easy dividing line between between those populations. So I agree with her that it's a little bit inherent to say, this group that we're defining as like white Europeans, we can study things within there, and that's fine. But some other group that might include more diversity, that's not fine. I mean, my sense of it is sometimes it's fine. And sometimes it isn't. And it depends on the question that you're asking. And there's going to be some level of confounding in any group above the family, and this is going to get bigger as you expand out. And there are ways to correct for it, but they're always somewhat imperfect. And I think that, yeah, this neat distinction between within group and between group just doesn't really exist philosophically or scientifically.
Yeah, one thing, one thing I pointed out, is, you know, and I pointed out like, my kids are of mixed heritage. So does that mean that they cannot use any PRS's from South Asians? Or Europeans? Because they don't fit into those ancestral categories in a neat or tidy way? Or are we going to just develop better methods? Or are we only going to have to use PRS's As in people who are f1 of South Asian and European background? Right? I mean, it's just like, if you take it to its logical conclusion, you kind of have all these little islands, these little genetic islands, but the reality is we are one species, there is gene flow, there is genetic variation, population structure is real. It isn't trivial, but neither doesn't mean that we're totally in commensurable, that we're totally incompatible. Right? So the issue with portability of GWAS, what do you think about it in terms of so genome wide association for the listener, which is basically the raw input to a lot of these estimates of heritability in a way like looking at these genes and figuring out how they add up or subtract. What do you think, you know, we've been talking about this for five years now Alicia Martin's paper, I think 2017 and American Journal of human genetics was super huge in this area, and I think they've actually there was an error in one of the packages and they've had to like reduce the effect. So last I checked they like the portability of GWAS rule of thumb to like from like Europeans to South Asians is like 75 80% The effect size, it drops to like 50 and East Asians and then it drops even further to Africans. I'm assuming that's gotten better. But I mean, do you think that this is going to be solved through the new technologies through brute forcing it, or through new methods?
I think it's gonna be a combination of new methods and more diverse data sets. I do think that genetics needs to get out of this mindset of only studying, so called within group things, which may has mostly meant defining some homogenous group of people with European genetic ancestry, and then just doing studies within that, and not worrying about anything else. But I think that it is hard, though it is actually hard to deal with the full genetic diversity of all humans in your study, because things become complicated when you have that level of structure and the data, you need to deal with that, and it makes the methodology difficult. I think the family methods can can help. You know, I know people, Patrick Turley, my colleague and the ssgc, he's doing really interesting work on this. Other people are doing interesting work, where I think, actually identifying the causal variance is going to be a big part of that find mapping causal variants, because probably, it's probably true that the the actual genetic architecture, the effects of genetic variants, are quite similar between, you know, different different people living in different places with different genetic ancestry is, but it's just hard to translate results, because we don't actually, we often don't actually identify the causal variants, we're relying on statistical correlations between the variants that we measure and the true causal variants. So I think that as the data becomes richer and more diverse and people develop methods is there's no reason why it should it should stay so Eurocentric, I think it's, it's a tractable problem, in my opinion.
Alright, so I want to I want to back up just for the listener, like what you said there, cuz I understand what you're saying. But it might, it might be a little confusing to them. So you know, what we've been talking about implicitly, a lot of the time is tagged snips, single nucleotide polymorphism variants that are associated with the functional with the mechanistic with the biological cause with the gene, the genetic position. And so if you have, say, 500,000, single nucleotide polymorphisms, within a genome, that is kind of tagging a region of the genome, because the these regions are co associated with each other, they're not independent. And the closer the region of the genome is, the more tight the correlation. So when you have a genetic position somewhere, you could have something I don't know, like 25 kilobases away, that is actually the causal variant, it might be an adjacent gene, or it might be the nearest gene to this intergenic tag. And that's actually what's causing the biologically significant characteristic. And so what you're alluding to here is, we can have a situation with a tag markers are somewhat different, because the different evolutionary history, the different, you know, population history of different populations, but the underlying causal variant or the causal gene or the causal region, could it actually be the same? Correct? Yeah, yeah, yeah. And so basically, if we have whole genome sequences, some of this will be obviated, some of this will be fixed, right? Because you don't have to rely on tag markers as much.
Yes, and I think also, just having more diverse data actually helps you pinpoint the causal variants as well, because you're not so beholden for the particular genetic history and what that causes in terms of correlations between genetic variants in one population. So actually, the solution in part is just to look in more diverse data that's just going to help us pinpoint causal variants better and build ultimately better models.
Well, so I mean, my final question that is actually like on my list, but I mean, it's kind of more open ended, like you know, it's when well when will behavior genomics be over? You know, when is this gonna like tap out in terms of at least like the statistical genetic results? I mean, I think mechanistically there's gonna be a lot of specific markers positions in genes to investigate for the bench people, the molecular people, you know, therapeutic people, medical people, but um, in terms of Okay, you got a big sample like I think the next next big paper is gonna have about 3 million individuals. I mean, you know, three, you know, 30 million, 3 billion like let's assume that it starts to get diverse at 30 million you know, what are we looking at here? Like you know, when do you anticipate like okay, you know, I mean, for example, they're not going to be doing too many more height GWAS's, correct. I mean, come on, like, how many more of those are you gonna do?
I think the common common snip effects on hight are like basically solved now. Yeah, I mean, I think there are obviously diminishing returns. And in terms of simply increasing the sample size of the current study design, going from 1 million to 3 million, so we, you know, this paper should be coming out soon, hopefully. And we do see a meaningful increase in predictive ability, our ability to predict people's educational attainment increasing the sample size from one to 3 million, it's going to, it's going to be less, you get less and less as you go out. So that that in that respect, just doing the standard study design within a homogenous European sample with more and more in the standard kind of genetic data. Sure, there's diminishing returns. So the area that I'm working in is, is the family data that is just getting started in my opinion, because that that's, that's something I think we need to do if we want to answer some of the trickier and more confounded questions, you know, a lot of what these critics say, the criticisms that people put to the field, I think a lot of that can be answered with with the family data, and we still don't have enough of that really, in my opinion, a sample size of 3 million of unrelated individuals, but for, for, for people with genotype parents is on the order of 10s of 1000s, not millions. So there's still a long way to go there.
Yeah, I mean, let me make it concrete for the listener, some of the criticisms and what I'm imagining. I mean, I mean, we've talked about this, I think, many years ago, this was obvious where it was gonna go, in terms of this family study, you can have a so let's say that, you know, let's say that there's an external physical characteristic, like beauty that you're positing is actually driving the impact in these GWAS's and beauty, you know, there's a small sample size, but you know, I mean, I saw something that, you know, attractiveness, you know, facial symmetry or whatever, like, by Raiders is like, you know, .65 heritable or something like that, which seems reasonable to me, that seems reasonable a number. So, you know, it's heritable itself, it's got genetic underpinnings. And so let's assume that you have a range of beauty within a bunch of siblings, they've all been rated. And then you have a range of genetic relatedness of all the siblings, and then you have a third trait, let's talk about EDU, I don't think that that should have been impacted by beauty. But this is just a thought experiment. Well, I mean, what you could do is, if beauty is driving this, you should be seeing the more beautiful siblings, you know, being more educated. And if you have like, mostly the genes, and then other random effects that are unrelated to beauty driving this, you should see that many, many beautiful siblings are, you know, they're less, they have less educational attainment, you know, perhaps like the polygenic risk score, if that's driving it, you'll see it within there. So you have these people raised in the same family with broadly the same genetic background, they vary on these characteristics, you probably can do GWAS for attractiveness. At some point, if people figure out a way to systematically if you can measure the genes, you can measure the genes that affect both of these characteristics. And you can track them and you can do complicated statistics, but you could probably actually do pretty simple statistics to see if the effects are happening the way that that you are predicting that they're happening, like we have the genomic data, we can have the phenotypic data. And so do you do people want to answer these questions? We could answer these questions, right? Like it's feasible.
Yeah, these these problems are tractable. And to be honest, I think we already kind of know that that's not what's going on. But we can more comprehensively answer them with more data and more data on families, more data on physical appearance on education, it's very much a tractable problem to see whether that kind of thing is driving these these signals, I guess. Yeah, like one thing related to that is, I think, you know, you might see these correlations between things like height and education in the population, and to some degree in the genetic results. But those kinds of things can kind of be generated by a sort of mazing or structure in the population. And some of the research I've been doing recently republished, but when you when you do the controls for the family, a lot of these kind of relationships between traits actually go away, which is kind of telling you the biology, some some of these things not really being driven by shared biology to being driven by some pattern of mating within the population where attractive people marry more educated people, you have some kind of structure in the population like that. That is social in origin, but if you if you do the properly controlled genetic study, you can remove that...
Yeah, we have the technology. So do we have the will? It's uncommon question. So I, you know, I basically asked you, I think mostly what I wanted to, to ask you, um, you know, at this point, I'm just wondering, like, Is there something that you want to say to the listener out there? Because, you know, I mean, I think like, I have a fair amount of listenership in the genetic space for people that are working in genetics. And some of these people are going to be, you know, they don't know your field, or they're skeptical of it. I mean, what would you say to, and you, yourself come out of like medical genetics and population genetics and math. So it's not like, you're a stranger to these people? What do you say to them? What would you say to them? Is there anything? Or have you said, I said, everything you need to say,
um, I think what I would say to them is that everything is kind of connected as well, and that the social sciences are not really walled off from the medical sciences, and things like your socio economic background, or your education, their predictive of health, and lifespan, and disease risk, addictive behaviors. You know, I think that it's a mistake for population geneticists and medical geneticists to wall themselves off from the social sciences, because they see it as controversial, they don't want to get involved. Because ultimately, all these things are connected, humans don't just do medical things and do social things, you know, that all related to each other. And we should study it all together. And I guess to people from the social sciences that are skeptical, I would also say that genetics is a potentially powerful tool to study social phenomena. So one social phenomena that I mentioned before is a sort of mating. So when people marry people who are like them in education, or height or whatever, genetics is a very powerful tool for measuring that. And economists and social scientists, some of them are really excited about the potential of genetics to get it. Old questions in the social sciences to understand things like, how does a policy change affect people differently, depending on you know, their characteristics, and that's something I've actually learned from working with with people in the SSG, I see that economists are actually very clever at finding these kind of natural experiments, things like a policy change, like oh, school leaving age, for suddenly change from being 16 to 18. And maybe that affected people differently, depending on that. Genetics, right? And, and genetics can be a tool that allows us to see more in this space and to study things more completely and wholly, rather than just saying, we're doing social science, we're gonna ignore genetics, or we're doing genetics, we're gonna ignore social sciences. That to me is not...
You're getting into Harden territory, you're gonna get you're gonna get in trouble. You better watch what you say. I mean, yeah, I mean, I think you're making some good points here. And it's hard not to draw policy implications when you study humans. And I mean, are you yourself? So I mean, look, a lot of these criticisms come from a place of fear. You know, there people are talking about eugenics. You know, I think you're you mentioned Patrick Turley, he co authored. I don't know, if you were on that piece of kind of against embryo selection. I mean, what is your take on all of these social fears? These like, secondary tertiary fears? I guess.
Yeah, I mean, I think the valid, I think that it's very easy to take genetics results, and use them to justify whatever policy you might like, and people on the bright do that they might justify, you know, social programs pointless because everything's heritable, so nothing's going to make any difference. I people like Paige might say, oh, genetics is a form of luck. And therefore, we need to redistribute to make up for people's lack of luck in life. Personally, I think people tend to see a justification for whatever ideology they already started with in the science. But I do also think that the policy should be informed by a true understanding of the world. But the relationship between genetic science and particular policies I think, is is not it's not particularly strong in my opinion, but it's people should understand the causes of differences between people and how they're going to interact with whatever educational or medical system that they end up in. But it's it's a very hard problem to do that well into into not just end up doing something that is gonna have unintended consequences, I think.
Yeah. Well, I mean, I really appreciate your time. I think we dug deep into a lot of questions that people had, if they're not satisfied with your answer they can read your papers if they don't want to read your papers that's on them you know I I have thought to respond, but I'm not in the field and you were much more eloquent and detailed about it. I think there are responses you can still be skeptical but this is not pseudoscience This is not built on nothing and you guys who are doing this work it's hard work and your your your motive is to understand the world as it is it's not it's not to be some eugenicist handmaid or something in the future. I mean, this is just ridiculous. So I really appreciate the time and you know, answering the questions and going into such detail I think it was needed it was necessary. And I think you know, I think a lot of people will get benefit from this podcast so thank you, Alex.