Alex Young: And I actually noticed, you know, these papers that came out around 2018, 2019 kind of time, academic papers looking at embryo selection, they had really quite, like, a negative framing about the technology. Okay. And when at least the titles did. And when I talked to people in private and when I did my own calculations, I kind of realized that this was not really accurate, that the negative framing was overly pessimistic about the potential of this technology, especially if the ability of genetic predictors would increase, which I, I knew it would, because as the, the size of the datasets gets bigger, the, the, generally the predictors become more powerful.

Steve Hsu: Welcome to Manifold. My guest today is Alex Young. Alex is a distinguished statistical geneticist, professor at UCLA. We've known each other for some time. I've wanted to get him on the show but, finally succeeded now.Alex is fighting cancer, which is something that we'll discuss, but I think we'll open with more scientific topics of broader interests, if that's okay with you, Alex.

Alex Young: Yeah, of course.

Steve Hsu: Great. So welcome to the show. Have been following your work for many years. We even submitted a couple of NIH grants together, although they weren't funded. The fault was surely mine, not yours. If I understand right, you were a math major as an undergrad, and then you did your PhD at Oxford under a, a a pretty famous geneticist.

Alex Young: Yeah. So I, I started off in, in math. I actually had a bit of an unusual
path maybe. I actually I got into I got admitted to Oxford for my undergrad, but I actually turned them down due to reasons of youthful stupidity. And, I ended up actually, like, dropping out of, of college in the first few weeks.

My mother forced me to get a job at a magic shop. So I went from going to Oxford to working, working in a magic shop and and I got fired from that job. So it was a bit of an inauspicious start to my academic career. And I ended up just emailing the the math department at Newcastle University, which is where, where I'm from, and they, they allowed me to join like halfway through the first year.
So had a bit of a, bit of a, a difficult start perhaps. But, but yeah, I did, did very well in math and got a summer project in statistical genetics working on yeast data actually. And I'd become very interested in evolutionary biology through reading, you know, Richard Dawkins and, and and, related kind of thinkers at the time.

I was kind of influenced by the new atheist movement, which got me into evolutionary biology. And I did this statistical genetics project, which went very well. It seemed like a very exciting field to me. This was back in like 20- 2009, 2010. So things like what, what, what was called then next generation sequencing.
It's still called that now for some reason, even though it's been around for about 20 years. But there was, there was this kind of sense that technological progress was creating a huge opportunity in this field, especially for people that had, you know, math and coding kind of skills like me. So instead of doing a PhD in math, I decided to sort of transfer over into, into computational biology.

So I went to, went to Cambridge and did, did a master's in computational biology with with Richard Durbin at the Sanger Institute. That was, that was pretty, pretty fun. And back then, actually, there was this, this, this problem, which is still around, called the problem of missing heritability, and that's sort of the problem that I first got stuck into during my master's.

And it's kind of stayed with me, with me ever since. And yeah, I ended up, you know, working on sort of mathematical theory in, in population genetics, heritability. I kind of rediscovered some, some old results . It's actually kind of interesting that I, I rediscovered these equations called the Kempthorne-Cockerham equations, which had been around since like the '50s.

And neither Richard Durbin nor Peter Donnelly-- So when I went to Oxford, I did my PhD with Peter Donnelly and, and, you know, Richard Durbin and Peter Donnelly are two of the, two of the most esteemed names in human genetics genomics, and neither of them were actually aware of these these classic results from the '50s, so they thought this was like a new thing.

But one thing I, one thing I actually realized from that is rediscovering old stuff is the best possible way to learn it. Because if you, if you rediscover something yourself, you really understand it.

Steve Hsu: Yep. Yep. Hey, I, I can't resist going back to your magic shop experience because when I was growing up as a kid, I was, I did stage magic and I'm curious, were you a magician as well as an employee at a magic shop?

Alex Young: No, I, I wasn't. It was basically my mother's friend. So my mother got me this job, and she ran this magic shop in, in Newcastle. And actually, a lot of the people who worked there other than me were magicians, so they taught me a few tricks, but they were, they were kind of like failed magicians though, which is quite a depressing thing to be.

So they were doing tricks that weren't particularly good. It didn't, it didn't make me want a career in magic I have to admit.

Steve Hsu: It's a, it's a tough career, but even to this day, I occasionally will go on YouTube and look at, I like sleights, sleight of hand, and just watch what people are doing. It's, it's, it's quite an intellectual challenge to reverse engineer tricks that other people are doing. So, yeah, it's pretty interesting.
Alex Young: Yeah. I, I, I've been to the Magic Castle in LA, actually. I don't know if you know about that place, but that, that was pretty cool.
Cool.

Steve Hsu: Okay, so now you're in graduate school, you're actually discovering things i- of importance, although maybe not the first person to discover them. And your timing is pretty good, right? Because wh- when did you finish your PhD?

Alex Young: In 2016.

Steve Hsu: Oh, yeah. So you, you finished right as UK Biobank was starting and deCODE was already chugging along. So there w- it wasn't this era where ... I always felt sorry for these evolutionary biologists because there was plenty of theory, but not enough data for a long time for these people.
And so you, you kind of came along at a period where there was a big jump in the amount of data available.

Alex Young: Yeah. So Peter Donnelly's research group, which I was in for my PhD, they were actually the main research group preparing the genetic data release for UK Biobank, the initial release. So I kind of had a, a bit of an advantage, I suppose, o- o- over other people in that I, I saw how that data was being processed and I think I was one of the first people actually to use the, the initial UK Biobank data release to study some, some scientific question.

So it, it definitely felt like a step change because before that, before UK Biobank, there was deCODE, which, which had a similar kind of sample size, the, like around 100,000 to 150,000, which, which was the sample size that the first UK Biobank release wasn't the full data. It was only about a hundred and fifty thousand.
But before that, you had like one of the big studies which, which Peter Donnelly's group had kinda led as well, was this Wellcome Trust Case-Control Consortium, and they were looking at samples, you know, in the thousands, not in the hundreds of thousands. And while they found some stuff it, it turned out that to really make progress, especially for these sort of complex traits and diseases that, that weren't amenable to, you know, more traditional linkage-based studies that people did, like looking for like rare, rare disease genes in families, you really needed those big samples.

So De- deCODE dominated for a while because they were the f- they were really the first to get the kinda scale that you needed to discover lots of disease-causing loci for complex diseases like, you know, type two diabetes and BMI and. And then UKB kinda completely revolutionized the field, I would say, and, and, and made it possible to not only a lot of the focus of genetic, human genetics in that time was finding, you know, strong loci with pretty strong effects on diseases that would tell us something about the biology. But then once you started getting to these sort of Biobank-scale sample sizes, that, that opened up the, the possibility of creating, you know, what we now call polygenic scores so genetic predictors based on the whole, the whole genome basically for traits and diseases which without sample sizes of, of sort of UK Biobank scale, that they tend not to really be powerful enough to have any useful applications.

Steve Hsu: Yeah. So you, I think, went to work at deCODE, right, after your, your PhD, and you must have lived in Reykjavik for, it was at least a couple years?

Alex Young: Yeah, yeah. I, I actually started visiting deCODE during my PhD. So yeah, Augie Kung, who I ended up doing my postdoc with, he came to visit. So actually, Peter Donnelly had been an advisor to deCODE for quite a while.

I think Kari Stefansson, the CEO and founder of deCODE, he'd wanted to recruit Peter in the early days of deCODE, but he didn't. He wanted to stay in Oxford or s- something. I don't, I don't know exactly why he didn't do it. But Augie ended up being the sort of main statistician of deCODE, and he came to visit.
And actually, this goes back to the you know, this work I actually started during my master's in, in Cambridge with Richard Durbin, which was on this missing heritability problem.

Alex Young: So maybe I should back up and say what the missing heritability problem is. So genetic association studies kind of promised to take us into a world where not only would we have some sense that there's a genetic influence on traits, but we could actually pinpoint which genes were, were important. And there was a lot of optimism in the early days of genetic association studies, GWAS, that we just have to collect, you know, a few thousand cases and controls for some disease, then we'd be able to explain all of the heritability that twin studies suggest is there for some disease like Type 2 diabetes or a trait like height, and it turned out not to be that easy.

So people would do an association study for height, and they'd find some variants in the genome that were associated with height. But if you created a, a predictor from those, that small number of variants, it would maybe explain, like, 5% of the variation in height. Whereas twin studies were saying that the heritability of height, how much of the total variation is explained by genetics, it's like 80%.

So this became labeled the problem of missing heritability. And back when I was starting out at Cambridge, there was this paper that got a lot of attention and and, and became quite influential, I think probably in part because Eric Lander was one of the authors. But it made this argument that non-additive genetic effects were, were the explanation for that.

They were leading to biased twin estimates that were too high, and we weren't detecting all of that heritability because we were just looking at the associations between genetic variants and phenotypes, and if actually it's due to these, like, complicated interactions than just the linear models that people were using would, wouldn't really work.

So I sort of rediscovered these classical equations from quantitative genetics that tell you about how different types of genetic interactions contribute to the overall, what we call broad sense heritability in the population. Then I generalized that to founder populations, so populations that have gone through some kind of bottleneck, so they, they have this kind of like background level of relatedness that changes the quantitative genetics.

And I thought, "Oh, I can go and apply this in the de-code data in Iceland," because that's a founder population. They have this sort of background relatedness, and it would enable me to measure the degree to which these more complicated kind of interactions between different parts of the genome are explaining variation in height or education or disease risk.

So I persuaded Augi Kong, that this was a good idea. Then I had to get Kari Stefansson's blessing as well, and If anyone knows anything about Kari, getting his blessing is, like, a highly non-trivial... That's actually a harder problem than the scientific problem often. So yeah, I met Kari and he gave me the blessing, so I started going to visit deCODE and, got to know Augi, and that ended up being very important for my career.
And I, I did my postdoc split between I was actually employed by deCODE for a while and lived in Reykjavik for about a year and yeah, worked with Augi for a few years in, in my postdoc.

Steve Hsu: So just so we have the scientific logic, clear, I, I remember this paper by Lander, and the co-author, is it Zuk?

Alex Young: Paul Zuk, yeah.

Steve Hsu: Paul Zuk, yeah. So I remember this paper, and I guess I wasn't I didn't believe it but was your idea that you might follow up on their paper using these new things you discovered in the deCODE data? So was it, was your motivation premised on them being actually correct?

Alex Young: Well, I, I guess it wasn't necessarily premised on it being correct. It was premised on it being a potential explanation and wanting to test it. And the theory that I developed indicated that the kind of data they had in deCODE, you would be able to actually measure these, the contribution of interactions using some of the theory I developed. Now, it actually turned out to be more complicated than I initially thought, as is often the case.

But that was what led me to, to Iceland basically, was, was that theory that I developed actually during my master's and in this missing heritability problem. But then kinda Augi, I think he saw that I had some ability, and he kind of manipulated me into what he thought was a more productive avenue of research, which, which he was right about.

Steve Hsu: That's what good advisors do.

Alex Young: Yeah.

Steve Hsu: So we'll come back to missing heritability because I think that's on our list of topics that we wanna talk about. But just for his sort of historical reasons, at the time, were you expecting that a big part of heritability would come from nonlinear effects? Are you surprised by how it's turned out?
Alex Young: I did think that interactions were probably more important than people had traditionally believed. I think maybe I wanted them to be because everything just being, like, this simple linear model kind of felt a little bit boring to me, and a lot of the math that you can work out for all of the kind of more complicated interaction models is actually

It's actually quite fun. Like, the, the, the theoretical side of it is, is pretty interesting, how it contributes to relative correlations. It shows you the way that The, the genetics world broadly is actually quite fractured.
A lot of the human genetics people actually don't know a lot of quantitative genetic theory. Whereas people who come from more of an animal breeding background, they tend to have a much stronger grounding in in, in quantitative genetic theory. And then you have the people that work more on model organisms like mice, yeast, whatever, and I'd done some yeast work.

And there was actually another paper that was published in Nature that got a lot of attention around that time, and it was looking at it was looking at yeast crosses and finding that sorry, my cat is trying to get involved in this podcast. Looking at, looking at yeast crosses between different yeast strains and seeing whether additive models or interactions could explain all of the, the heritability of the, the growth of these different yeast traits and different media.

And I actually applied some of the theory I developed to some of those data, and it showed that not only pairwise interactions, but actually these really complicated like third and higher order interactions were, were quite important for explaining the, the heritability of, of growth traits in yeast.

And this, this is something that we see quite often, that these non-additive effects are actually quite important in, in lab experiments and model organisms. So people who come from that world tend to think, "Oh biology is this really complicated nonlinear thing," which, which it is in many ways, and therefore human genet human genetics is missing out on important things by ignoring interactions.

But then the people that come from maybe the, the sort of animal breeding, quantitative genetics background they tend to be more skeptical of the role of interactions. And then, you know, it's, I it's like they, they never really talk to each other. I still think there is a bit of a question about the role of interactions and missing heritability.

There's still some, some odd results that maybe the interactions can explain. But certainly for things like height, you know, the additive model seems to explain pretty much all of the, the heritability, and exactly why that is can be quite complicated to to explain. I, I guess I was maybe a little disappointed that it turned out to be simpler than I thought, but I've accepted it now.

Steve Hsu: When I came into the field, I was very agnostic about this point and how it would turn out, 'cause I was aware of both perspectives. And a lot of my, The early stuff that I l I worked on in looking at different algorithms that would be useful for building polygenic predictors. I, I considered both the possibility that things were largely additive, and also the possibility that there were big chunks of non-linear stuff in there.

I... By the way, speaking of yeast work, I always wanted to ask you about this. I think one of the guys I went to grad school with is one of your colleagues at UCLA, and works on yeast. Is it Leonid Kruglyak? Do you know this guy? Oh, yes. I

Alex Young: think He's the, he's the chair of my, my department, actually.

He's my chair. Yeah. Yeah.

Steve Hsu: Yeah,

Alex Young: so he started- Yeah, yeah. No, so actually, I, I came across Leonid's work when I was an undergrad still, and he was actually somewhat aware, I think, of my yeast work as well when, when I eventually met him. Yeah he wo he worked with Eric Lander on some of these foundational algorithms and hidden Markov model algorithms as well back in the early days.

Steve Hsu: Yeah. When, when he and I were in graduate school, we didn't have a single hu we were far from even a single human genome. We were, you know, this was very early days. And he worked with a guy called Bill Bialik in biophysics. And Mm so then, yeah, you end up, I think, post-docing with Lander after that. Yeah, so somewhat a small world.
So so
Alex Young: I think it was actually hi was it, I think it was his group's paper that, that, that showed that these interactions were, were important in the yeast growth traits. Yeah. You know?

Steve Hsu: Yeah, that's how I remember it. So okay from deCODE you somehow made your way to the West Coast of the United States and you got involved withsocial science, a, a kind of set of phenotypes which are sort of in ill repute among some geneticists. Do you wanna, do you wanna mention that a little bit?

Alex Young: Yes. Well, I, I blame Augie Kung entirely for that, actually. It is his fault in a way that, that I went down this path, 'cause before, before that, before I met Augie, I wasn't really that I'd always been interested in these slightly more controversial topics like why is why do some people get more education than others? Why are some people smarter than others? But I hadn't, it hadn't really been a focus of my research. Actually, during my PhD with Peter Donnelly, I'd been working on much more standard sort of biomedical human genetics questions, like looking at BMI.

I, I was looking at developing methods for figuring out gene-environment interactions affecting BMI, which they do s like, that's one kind of interaction that does seem to be real. Like, the, the, the, the environment does interact with our genetics to affect BMI. So that was kind, the kind of stuff I was doing.

And then- Once I started working with Augie, he'd been collaborating actually with the SSGAC, which I ended up joining. So that's the Social Science Genetic Association Consortium, which had actually been set up by a group of economists, including Dan Benjamin, who I work closely with at UCLA, and David Cesarini at, at NYU.

So they'd been conducting these genome-wide association studies of education, which a lot of people didn't really want to happen, and they've started there's actually some interesting stories about the early days of that, how, you know, they had collaborators who then pulled out after they found stuff.
Like, they, they, they didn't want they were willing to contribute the data to these genetic association studies of education, but then when they found genetic variants that were robustly associated with education, they didn't wanna be associated with that. So it was quite a controversial thing, and and Augie got interested in it, and I think Cary Stefansson, you know, his background actually was in neurology, so he's always been very interested in, you know, genetic effects on, on, on brain behavior, psychiatric phenotypes.

So they were quite interested in and they had education data in, in, in the decode data as well. They had one of the biggest samples. I think maybe it was the biggest sample back then. I'm not sure. So they'd contributed actually to, to some of the earlier EA educational attainment GWASs. And Augie basically found this very curious result actually, which is how he got me into this whole area.

So the SSGAC had been constructing these polygenic scores for educational attainment, so genetic predictors based on the association results, adding up loads of different variants across the genome. And until Augie did this, no one had really thought about it that deeply. I think the SSGAC had thought about it a bit, but maybe hadn't had quite the right data to look at it.

But Augie basically looked at whether the genetic predictor worked in families as well as in the population. So you can, you can look at the association between the genetic predictor for education and educational attainment just in a sample of unrelated individuals, and you get some correlation or R squared or whatever.
Then what Augie did is also check this within families. So controlling for the, the genotypes of the parents, then you're just using the random variation in genetic material in a family that derives from meiosis. So whether you inherit one or the other copy of, of of each bit of the genome from each of your mother and the father, it's like a random coin toss that's sort of independent of environment.

And when Augie did that, he got this result that- I think was quite surprising at the time and got a lot of attention. We published it. I was on this paper, although I only played a relatively minor role, but we published it in Science eventually in 2018, although the results had been around for a while.

The, the association roughly halved when you, when you looked at it within family compared to in the population. And figuring out exactly why that is has kind of occupied a lot of my research since then. And Augie kind of encouraged me to think about, you know, how this family data would also, could also be used to estimate heritability in a more robust way.

So that's what, that's what led me onto, onto one of my contributions to the field, which was this method for estimating heritability that uses random variation and relatedness within, within families to remove, bias from heritability estimates. So, so that's sort of what led, led me into doing social science genetics, was the work I did at Decode with Augie exploiting family data and looking at education, and then that got the attention of Dan Benjamin and the SSGC, and they basically recruited me to the US.

Although I actually, I nearly went to China before that. So the, the I don't know if you know about this, but the BGI tried to recruit me, and I, I very nearly went, actually. I spent a week out in in the China National Gene Bank in Shenzhen, and they made me a very good offer. I actually resigned from my my Oxford post-doc and started learning Mandarin, so I nearly went there, but that's, that's another story.

Steve Hsu: Oh, what a small world. When, when, when I first got started, I I did a bunch of work with BGI, and one of the guys that I worked with is a guy called James Leigh, who later also worked with SSGC. And, and I remember Yeah,

Alex Young: I know James.

Steve Hsu: Yes, he was transitioning between the BGI work that we were doing and SSGC. I, I used to argue with, not argue with him, but we used to debate whether EA was a good educational attainment was a good phenotype. And I actually thought within family, EA would, you know, a lot of the variation EA would just sort of disappear because it was very influenced by the family.

So we, we actually talked about this a long time ago. And, yeah, so it's, it's a very small world.

Alex Young: It is a small world once, once you get into the sta the statistical genetics world isn't that big, really. There's only so many places that are very strong in, in it, actually.

Steve Hsu: So let's, let's talk about, I think, the, in the, in the list of I want, I wanted to I'm really interested actually in your intellectual history 'cause I, I, I do admire your work. And so I wanted to go into that at some length before we started the topics that, that you and I discussed over email. But maybe we could switch over to that now.

Alex Young: Sure.

Steve Hsu: The first topic you had was about embryo selection, and that isn't something you were involved in for a long time, but recently you've gotten involved in it. So maybe tell us, c- in a way that's sort of like jumping, you know, I don't wanna say the dark side, but it's jumping to the, one of the more controversial applied areas of this work and maybe you can just tell us what kind of finally got you there and what you find interesting in that field.

Alex Young: I became interested in the embryo selection topic. I think that was probably, yeah, around the time I was joining the not because I joined the SSGC, but around that time there were a couple of papers that came out about the efficacy of embryo selection, like, one of, one of which was partially is from the SSGC. I think it was in The New England Journal.

And then there was also another one from Shai Carmi. I read those papers and I was pretty interested in the basic idea, and it actually seemed very closely related to a lot of the family, because it's basically looking at prediction within a family. So I, I kinda realized that a lot of what I was doing about exploiting the properties of family data to improve inferences and, and understand how polygenic prediction is working, heritability, things like that, that that could really have a direct application in this domain.

And I actually noticed, you know, these papers that came out around 2018, 2019 kind of time, academic papers looking at embryo selection, they had really quite, like, a negative framing about the technology. Okay. And when at least the titles did. And when I talked to people in private and when I did my own calculations, I kind of realized that this was not really accurate, that the negative framing was overly pessimistic about the potential of this technology, especially if the ability of genetic predictors would increase, which I, I knew it would, because as the, the size of the datasets gets bigger, the, the, generally the predictors become more powerful.

So I, I, I definitely was interested in this space, but I didn't really get involved with it until 2023. I'd, I'd, I'd kind of made a decision that I'd like to get involved with it, but I never directly did until yeah, I was, I was approached. I mean, it was sort of like, what could be my contribution as well, because it wasn't really of that much interest to me to just try and compute better embedded polygenic scores other people were working on that and doing, but, you know, that wasn't really my specialty.

But in the end, so Herasite, which were operating in stealth mode at the time, they, they approached me with quite an intriguing idea, basically, where, you know, could we design an algorithm that would turn a routine test for aneuploidy, so chromosome an anomalies like Down syndrome, could we turn the data for that, which is, like, really crappy data, it's, it's super sparse data on the genome.

Could we turn that into a comprehensive genome profile on the embryos? So that was sort of a scientifically interesting idea to me, and I had some expertise in the kind of algorithms that are needed to do that, so that's why I ended up getting, getting involved and, you know. Yeah, it definitely has cost me career-wise, to be honest.

But I felt like there was, there was, there was sort of a fake consensus that had been generated in academic human genetics that this technology was, you know, not only maybe ethically questionable, but actually also just ineffective. And, you know, the smart people knew that that wasn't really true in private, but everyone was afraid of saying anything positive about this technology in public.

And I think even to this date, I'm basically, like, the only, pretty much the only, like, well-known person in, in the human genetics world who's come out as pro this technology. Even though I know quite a few pe- people in private who are pro it, but they, they don't wanna suffer the, the career consequences I have.

Steve Hsu: Yeah, well, so I'm obviously I'm super interested in this since I'm one of the founders of one of the companies in this space.

Alex Young: Yeah

Steve Hsu: so I remember the New England Journal article that, you're on that paper, are you, or,

Alex Young: I'm not, no, no.

Steve Hsu: It's Benjamin and Ceccarelli and those guys.

Alex Young: Yeah, Patrick Turley, I think he's the first author. So yeah, there's s- Yeah ... there's quite a few, but it was from before I joined the SSGC, that paper. Okay.

Steve Hsu: Yeah, so, so they and Shai Karmi wrote papers which were, I think, scientifically okay in the actual content of the paper, but as you, as you point out, the title and framing of the results is, is very negative, and it's sort of contradicted by what's actually in the paper, but nobody reads the actual paper.
Right. They just yeah. So I don't feel like there was a real scientific controversy, like, a controversy at that time. It was all about just, them not wanting to be associated with this because many people are against it, I think primarily on ethical grounds. And n but I didn't realize that your getting involved in Herasite had actually already hurt you a- in your academic career. So I guess are there people who won't collaborate w with you because you're a part of the Herasite team now? Is that the situation?

Alex Young: Well, you know, I guess maybe jumping ahead a little bit to the, the whole cancer situation, but basically I, I kinda helped launch the, the, the company publicly on, on Twitter X in, I think it was late August last year, while I was actually in the middle of chemotherapy infusions and I, I wouldn't recommend launching a controversial startup and going, going viral.

I think my tweet got, like, 1.6 million views or something. Going viral on, on X for a controversial startup while you're in the middle of chemotherapy, that was, that was an intense week. Actually, it was also the week that some of the grant funding I have at UCLA kinda got temporarily paused by the Trump administration.

So all of that happened in one week, so that was kind of intense. And then following that, yes, I had a couple of collaborators pull out of a paper that I'd led as sort of a big meta-analysis paper doing a family-based version of GWAS, which is a sort of methodology that I've, developed. and it's quite important for various, topics that we might discuss.

They actually accused me of doing something illegal which, which was just completely spurious. And pulled out of the paper, pulled the data out and yeah. That's, that's been, been difficult and I also basically I lost a job offer as well. I had a faculty associate professor offer at a R1 US university that was then rescinded following that coming out and I think.

Well, you know, this is kind of the even actually before any of the embryo selection stuff came out I'd already suffered career con consequences just for maybe talking to, like, the wrong people. I think the thing that maybe got me into the most trouble was going on Richard and Anya's podcast and maybe saying some things on, on Twitter that weren't completely aligned with various orthodoxies in the field.

And there is actually there's a remarkably tight degree of ideological control in the human genetics world and I think it's partly because it touches on various verboten topics and people under it's, it's, it's interest it's an interesting social psychology phenomenon but it's like it's never really ex it's often not really explicitly said, "Oh, you can't do this, you can't do that," but everyone understands that there's, there's kind of an orthodoxy that if you, if you step out of line publicly Then it's gonna cause you problems in terms of you won't necessarily get canceled, but you won't get that call back for that job interview, or you won't get that collaboration, you won't get access to that data.

And actually, because of success in the field of human genetics depends pretty strongly on data access actually. That, that I think is part of, part of the mechanism th through which this sort of orthodoxy is, is enforced. So there's, you know, a lot of doing large scale human genetic studies is gathering, gathering a lot of data from a lot of different people. So you have to make sure everyone is your friend, you

Steve Hsu: know?

And then, you know, I think the, obviously the third rail is talking about heritability of cognitive ability and, and then within that, obviously the question of whether there are group differences, so whether different groups have different polygenic averages Mm-hmm for traits like that.

Obviously it's super sensitive. And although I did get into inter- I did get into embryo selection, I've always tried to avoid the cognitive ability thing, just because, for business reasons, we didn't it, it was just too controversial for the clinics, the IVF clinic. I mean, we we now work with hundreds of IVF clinics around the world, our company.

But that particular phenotype is something that's still, there's huge amount of sensitivity to. It's not a health or disease, directly a health or disease trait. And so it's, it's, it's kind of too sensitive even for our research group to work very much on. But Herasite- Yeah seems to have jumped completely in, into this now. And so now it's, it's really out there. Like people can go get their embryos evaluated for cognitive ability.

Alex Young: Yes. Well even before I got involved with Herasite, they'd focused quite a lot of their efforts on building a powerful genetic predictor for IQ. I think that was maybe where they saw part of the gap in the, in the market, was that other companies weren't actually doing that.

And there is actually a lot of demand for that. Although as you say, it's true that quite a lot of IVF clinics are not comfortable with like offering it themselves or advertising it. And I mean. That's actually part of the part of the reason for the development of this, this algorithm that I developed for Herasite was in a way to kinda try and put power back in the hands of the, of the couples undergoing IVF rather than the clinics.

Because people can, you know, access their own data from this PGTA test, which is pretty standard, and then that can be used for, for creating an IQ predictor even if the clinic doesn't want you to, to do that. Although that, yeah, can cause issues. But yeah, I mean, Herasite, they I wasn't involved in this work, but they, they put together a, a much more powerful genetic predictor of IQ than I thought was likely to be possible in, in, at that, at that moment with, with the available data.

And, and I think that made a, a much, much more worthwhi made it more worthwhile to take on the, the risk of, of offering that because if you can offer something, if you wanna, if you're gonna do something controversial, you want it to at least be, you know, effective and useful so that there's some payoff for taking on the controversy.

And while I think their estimate now is that if you're like a if you're a European ancestry couple with, like, 20 embryos, you could boost your latent, the latent, sort of, general ability factor by up to about nine points on average. So that, that kind of level of increased in so that's just an average, of course.

Like, that's another thing I think that's often missed in both the academic literature on embryo selection and in the broader discussion of it, is people focus on these sort of average gains for an average randomly sampled couple or something. But in reality, it depends on, you know, the particular set of embryos you have.

You can have a much wider range than the, than the average, or you can have a much more narrow range than the average. But I think that kind of level of effect starts becoming, you know, meaningful in the- Yeah the population of parents that did IQ selection, that would be, like, a noticeable difference in average Yeah cognitive ability from the population that don't.

Steve Hsu: Just for the audience, let me just recap 'cause I've not, I, I think a lot of people are not really, up to speed on what you were just discussing. So if you use this improved, IQ predictor that Herasite is offering, according to their simulations, if you, let's say you have 10 or 20 embryos, you could have an average gain of something like 5 or 10 IQ points from using their technology versus, say, random selection among those, say, 10 or 20 embryos.

That's the kind of ballpark, and that's getting to a point where it's pretty significant. Yeah and I think your work so just a little background for the audience. In the US, it's Pretty much standard of care for there to be at least some kind of weak genotyping or genetic screening of all of the embryos produced in the US. So standard of care would be biopsy is taken. They do a screen called PGT-A. So PGT is pre-implantation genetic testing, hyphen A means for aneuploidy, which means you're looking for either an extra copy of a chromosome, like which would cause Down syndrome or some gross rearrangement of the chromosomes.
And for that, you don't really do a high resolu you don't need to have necessarily a high resolution genotype of the embryo. You get a very noisy, I think as you said, Alex sparse sampling of

Alex Young: the

Steve Hsu: different things.

Alex Young: It's only like two, two to six in 1,000 positions in the genome are, are read basically.

Steve Hsu: Yeah.

Alex Young: Yeah.

Steve Hsu: But nevertheless, if you have the genome of mom and dad, I think you developed a method where with the sparse data from each embryo, you could figure out which chunks of DNA you got, the embryo got from mom and dad, and then thereby reconstruct the polygenic scores even though you don't have high quality data from each embryo. Is that, is that a fair summary of what you did?

Alex Young: Yeah. Yeah, no, it was, it was kind of surprising how well it works actually, because I was just saying that the data they generate so actually some of the PGT-A data, they use an array type technology, which is similar to what Genomic Prediction use. That's actually a much better technology for, for doing genotype imputation and figuring out the whole genome of the embryo.

Whereas what they, what's become more common now for this PGT-A test is this very, very sparse kind of shotgun sequencing, which just randomly kind of samples a, a small number of sites from, from the genome. And then that can be sufficient to detect, say, an extra chromosome. But using that to reconstruct the whole genome, I was actually skeptical that it would work when I was first approached, and we ended up coming up with a pretty innovative algorithm that, that works surprisingly well for it.

So yeah, it, it's all you need for in theory, all you need is that, that sparse PGT-A data for, to get these polygenic predictors very accurately. And if you wanna get de novo mutations, then you, that's, that's a different matter. But apart from de novo mutations, you can actually do very well just with this PG, this PGT-A data on the embryos and the sequences of the parents.

Steve Hsu: So in your case for your method though, do you need whole genome sequences for mom and dad?

Alex Young: you, I mean, that's the way that we've, where Herasite is doing it. I, I don't think you would actually technically need that. So you could actually use array data on the, on the, on the parents and then Impute that from a reference panel and then impute that down into the into the offspring.

It wouldn't work for, you know, part of, part of what we wanted this to work to, to genotype rare disease-causing mutations as well. So if you wanted to detect them in the parents, you need to do the high cover, high quality whole genome sequencing on the parents. If you were just gonna use like the cheaper or simpler kind of array genotype on the parents then you wouldn't be able to get at the you know, these rare pathogenic variants that the parents might carry as well.

Steve Hsu: And just to complete the thought for the audience. So because PGT-A is offered by pretty much every clinic in the United States, and it's actually st- pretty much standard of care now and patients have the legal right to download that data. So once they

Alex Young: Under HIPAA, yeah.

Steve Hsu: Yeah, so once they go through the standard procedure, which everybody in the US, not, not everywhere in the world, but in the US everybody pretty much goes through this if they're doing IVF. That noisy PGT-A data can be used by Herasite to reconstruct the genome of, or at least the genome well enough to compute the polygenic scores for each of the embryos.

Alex Young: Yeah, and it's actually used in other countries as well. Fun- funnily enough, I discovered PGT-A testing is actually illegal in some European countries, which was kind of shocking to me. Like some of the Scandinavian countries it's illegal to test for, you know, Down syndrome before implantation, but it's legal to abort after implantation if the fetus is found to, to have Down's diagnosis. Which seems an odd way to think about things ethically to me. But it's, PGT-A is legal in a lot more countries than em than embryo selection is.

So actually for example, there's a clinic in the U in London, an IVF clinic in London that's been you know, some of their customers have got their PGT-A data and sent it to Herasite and then you can actually like serve customers in jurisdictions where, you know, the regulation is technically stricter but because of GDP so GDPR is a bit like HIPAA in the US and that gives people the right to their PGT-A data and then. I mean, it, I have to say, maybe, maybe I'm just more libertarian than the average person, but, but it seems, it seems strange to me that you shouldn't be allowed to look at the genetic data on your own embryos, personally. Yeah. And that's what a lot of the, the regulators seem to think in European countries.

Steve Hsu: The regulatory situation varies by country obviously. The US is quite

Alex Young: Yeah

Steve Hsu: liberal. But it's also true in other countries that even if it's allowed, not everybody gets PGT-A. So, Yeah but in the US it's more

Alex Young: common It's more common in the US, yeah

Steve Hsu: You've done really impactful work for Herasite. I think this is an important advance. And d- I, I guess you intend to continue working with them.

So it's not, it's not that these the people who are against you in the acade in academia have scared you away from working with them.

Alex Young: Well, it's one of those things where I kind of knew there would be blowback. I d- actually didn't expect collaborators to pull out of a paper that, you know, that project had been going on from before Herasite had ever been founded or anything, you know?

And then the, they pull out at some later date. I actually didn't expect that because I felt like, well, you know, a lot of people in human genetics have financial interests in, in associations with pharmaceutical companies or other kinds of biotech companies. So I didn't think it would be, like, an issue that a collaborator has some pretty unrelated outside industry association because that's very common.

In, in some sense the blowback was, was less than I expected. I mean, there are some very ideological people. Tends to more come from the pop gen world for some reason. They tend to be more ideological than the, the bio- the biomedical people are a bit more pragmatic-minded, I think.

But the, the population genetic theorist people, some of them are very ideological I think, and they're very against this stuff I mean, I still have my job, so.

Steve Hsu: So I, I don't wanna dwell too much on the controversy point, but I am curious whether, you know, given that, you know, we're both scientists and we're sort of sworn to, or at least internally sworn to pursue truth and, and deeper understanding, did anybody actually come to you and say, "Hey Alex, I've worked with you for years on this project and or, or on other papers and I respect you intellectually and I just I just can't agree with your developing this algorithm for Herasite, and therefore I can't work with you." Did anybody actually just lay it out like that or?

Alex Young: No. No. Not, not really. I mean, it's... The way that these things work, it's not like a very in your face. I mean, yeah, some people did say some, some kinda unpleasant things on, on Twitter and Blue Sky, but they weren't necessarily the people that were actually causing me directly any, any kind of difficulties. So it's, it's, it's the way that these things work, I guess the sort of the, the big pr high profile, full-on cancellation kind of things are different.

I'd say the way things work is more the It's all done kinda like in a bit of a passive aggressive kinda way. It's like, okay, now you're on the outside, you know?

Steve Hsu: Yes.

Alex Young: It's more like high school mean girls kinda thing.

Steve Hsu: Yeah.

Alex Young: Than, than You're in the

Steve Hsu: out

Alex Young: group than like someone directly coming to you saying actually, you know, that's so James Leigh did actually directly say to me, he's like, "I really respect you as a scientist, but I'm totally against what you're doing ethically." I mean, he has his own ... He's, he's a Catholic and, and, and, and some of the religious people are very opposed to, to IVF in general, and then embryo selection as, as a part of IVF. So some of them actually were a bit more direct and I, I actually find that objection, although, you know, I'm, I'm not, not a Christian or, or a Catholic or whatever, but I kinda like understa It, it's, it's sort of rational given their premises. But then the other people that are opposed to it, their reasoning is often seems to me quite incoherent actually.

It's more like they just don't I think they just don't wanna be associated with it. Yeah. Because it can l- be labeled eugenics, and people don't wanna Yeah ... be labeled with that label. I think that's the underlying psychology of it, more than some Yeah I mean, there's some people worry about inequality. I think that's, you know, a legitimate point that you can worry about and people could debate. But should be something people a bit like tax rates or something, like, should be something that, you know, reasonable people can, like, disagree about.

Steve Hsu: Yeah. In the case of James, whom I've known gosh, since around 2008 or something, or even earlier than that, so it's a l a long time and, and in a way I learned a lot of this subject from him because when I first got into it he, he knew a lot more about it than I did.

His beliefs evolved in this over time. Like he was not opposed to embryo selection when we first started working together.

Steve Hsu: Yeah but then I think a lot of it had to do with the birth of his kids and, and somehow his feelings changed and he is just on moral grounds, against embryo selection now.

And yeah so I, I agree with you though. His views are totally coherent given his basic premise. Whereas someone getting mad at you, I mean, the, the stuff that your algorithm is used for, it might be used to, you know, help select against some very bad polygenic risk. It's nothing it doesn't even have to be applied for IQ.

It could be some disease risk that someone can just now they can use this noisy PGTA data to ensure that their daughter doesn't have high breast cancer risk or something. And, like, really? Is that that objectionable? I just, it just seems very crazy to me, but.

Alex Young: Well, I mean, I, I agree with you that, you know, the, the you know, the risk reductions can really be quite substantial, especially for fam families with a family history of disease.

Like You know, you can, you can like half your risk of passing on breast cancer to your daughter, for example, or something if you do this kind of selection. I mean, this if this if, if there was a drug with that kind of effect, we'd be calling it a wonder drug. Yeah, exactly. But the, the general vibe about it is, is super negative.

And, and it seems it, it's something I've, I've said before, but human genetics can be a bit of a curious field that's afraid of being too useful. Like, people don't, people don't want it to be applied for embryo selection or gene editing. They don't want certain questions about the role of genetics in socioeconomic or like racial inequalities to be answered.

And really, these are some of the most important applications of genetics, in my opinion. But the field is kind of against, against those applications.

Steve Hsu: Yeah. I, I think if it's too effective, then it, it borders on what they might consider eugenics, and so they don't want it to work so well.

Alex Young: Right. Yeah, yeah. I mean, there's certainly people that want to kinda minimize the efficacy or, or, or utility of genetics in, in, in these areas.

And and there are also some people that maybe, you know, you get on X that takes I guess it's, it's one of those topics where it does kind of, it does touch on people's political and moral intuitions about human society. And people bring those priors into, into this. And, and if you have a prior that everyone should be kind of equal, then you're gonna feel more or less uncomfortable about certain things.

If you have a prior that it goes in the opposite direction, then maybe you, you want it, you want genetics to explain all socioeconomic inequalities, which I think is also, you know, a bit of an extreme position. But my position is more that we should try and do do these studies objectively and rigorously using the best methods and data, but that seems to be quite a minority position unfortunately.

Steve Hsu: Yeah. It's, it's kinda shocking to me as, as someone who's, you know, definitely in the out group coming from a different area of science.

Steve Hsu: So one of the things you, you wanted to talk about was your paper with Peter Visscher which uses this Mexico. Is it Mexico City Project or MCP? I forgot what the acronym is.

Alex Young: Mexico City Perspective Study. MCPS, yeah.

Steve Hsu: Right. And so this study, because there's a lot of admixture in, in Mexico between people with sort of indigenous heritage and more European, say, Spanish heritage sometimes in a family you get a kid who has bigger chunk, just through random recombination, has bigger chunk of, say, indigenous genetic type or more regions of indigenous type, than their sibling. And I think in that study- Mm-hmm ... then you were able to study in a way group differences, right? So

Alex Young: Yeah

Steve Hsu: If, if one of the siblings by luck of the recombination had a genome that was more kind of like in the indigenous population, they tended to be less tall than their sibling even though they were raised in the same family. And, and also I think they had elevated diabetes risk. Is, is that, is that roughly the kind of thing that was in that paper?

Alex Young: Yes. Yes. So I mean, there's been so studying the genetic contribution to group differences that's probably the most verboten topic of all. And I think for example, if, if you take the training module you need to access the, the NIH's flagship data set, All of Us, it kind of tells you, "You're not allowed to do that." But I think this taboo maybe is breaking, you know, with this, this paper, this Mexico City paper, which I'm, I'm a co-author of. There's also this selection paper from David Reich's group. But so, you know, studying the genetic contribution to group difference is actually, it's, it's, it's a, it's a tricky problem and we haven't really had the data to really, I would say estimate the contribution in, in a, in a well-controlled way.

So, you know, the basic idea of an admixture study is just that you look at the correlation between so in Mexico City, I think it's about two-thirds of the ancestry is Indigenous American Mm-hmm probably primarily like Aztec origin. And then most of the rest is, is European origin. So you could just correlate how much indigenous versus European ancestry an individual has with their height, their diabetes risk, how far they go in school.

But that would in theory at least be confounded with other factors. You know, people with more European ancestry might be from better neighborhoods, wealthier families, descended from the conquistadors or whatever. So it's possible that this kind of this ancestry environment correlation could be driving any association you see between global ancestry proportions and traits.

So this idea to do it within family, it's been around for a while, but there hasn't really been the power in any data set to do it until this Mexico City study got, got genotyped and became available. I think Regeneron actually funded a lot of the genetic data generation. So, so in this study we looked at, like, there was about 17,000 families and looked across a bunch of different traits.

And so, so instead of just looking at the correlation with ancestry, we look at the correlation with ancestry within a family. So basically like whether one sibling has a, has more indigenous American ancestry than the others. And, and, and what correlation does that have with the trait? And that basically is a, you're estimating in some sense a causal effect of differences in, in ancestry.

Although there is this famous thought experiment from Yanks, the red hair gene thought experiment. So it's a causal effect of ancestry, but it doesn't tell you how it's mediated. Though it possibly could be mediated through a social mechanism like ancestry related appearance discrimination. So that's an important caveat that you can't completely rule.

I mean, I think there are ways you can test that kind of hypothesis, but we didn't really get into that in this paper. But it was, the, the results were very interesting nonetheless. So if you, if you just look at the correlation with ancestry, you see that the, the correlation is equivalent to saying that there's about a two SD difference in height between 100% indigenous American, 100% European ancestry, and there's also about a two SD difference in educational attainment, how many years of education someone gets. So they're actually pretty large differences. Also higher type 2 diabetes risk for people with more indigenous American ancestry.

Steve Hsu: Just to clarify that, that the numbers Yeah you're giving me are just now are not within family, right? It, the, so

Alex Young: No

Steve Hsu: this is in the

Alex Young: general population. No, no. Yeah. Yeah. Yeah.

So these aren't within family, so they can be confounded. Then when we look within family, it appears that basically the, the whole height difference, the whole type 2 diabetes risk difference, that's genetically caused, caused by ancestry, inherited ancestry differences. So it doesn't seem to be related to the environment really.

So that in and of itself is pretty interesting. I mean, heights, like, I mean, you, you probably expect that height differences between populations are at least partly genetic, although even that is controversial to some people. But interestingly enough, for education, the effect goes to zero within family.
So this was actually sort of the politically correct direction for the results.

Steve Hsu: On the, on the EA result, it's within family, it's statistically consistent with zero, right?

Alex Young: The point estimate is basically zero. I mean, the, the, the confidence interval is still somewhat wide, but it seems like the association we see in the population between ancestry and education, it's not driven by the, the yeah the causal genetic effects. It seems like that

Steve Hsu: Yeah

Alex Young: with some decent Fair level of precision. Now, you know, this educational attainment phenotype in Mexico City appears to be generally not very heritable, so that might be part of it, but it's an interesting result nonetheless. I a strong hereditarian hypothesis would have predicted that this was genetically explained. If you, if you think that basically all socioeconomic outcomes between groups are genetically caused differences are all genetically caused, it doesn't necessarily translate to other things we might be able to study with this design if we had the data.

Alex Young: And I mean, I think a, a broader point beyond the results of this paper though is that, you know, these questions are actually tractable with the right data and the right methods.

But people kind of don't want them to be and, and all there's, there's been kind of a push not to especially in the US actually, and in NIH datasets where you've basically been blocked from studying the genetics that... James Leigh, he, he wrote an editorial about this. You've basically been, been blocked from studying things like the genetics of education, IQ, even BMI and alcoholism in some datasets because they're deemed stigmatizing.

And in NIH, all of us, you have to agree not to study the genetics of group differences. So are we saying that, what are we saying about that? That it's, it's not, it's not possible? I mean, this, this study is showing you can, you can study it in a, in a, in a pretty rigorous design and get some interesting results, so we, we're saying that's not allowed in, in the US?

Steve Hsu: Yeah, it's crazy. I would have thought that Trumpers would have, like struck those rules at NIH by now, but maybe they haven't gotten around to it.

Alex Young: Yeah. I mean, I heard that there were plans to, to do that, but as far as I'm aware, it hasn't actually been changed. Probably takes some bureaucratic machinery to be changed within the NIH, 'cause actually the way that this is controlled is actually quite complicated and, and depends on the sort of NIH org chart.

There hasn't been a it's not like an NIH-wide policy that the director can just change, I think. It's, like, something decided by particular data access committees to control particular datasets within different sub-components of the NIH. So, you know, but that's actually another kind of point that the sort of data bureaucracy has got pretty out of control in human genetics and, like, people seem to want to make it worse.

Steve Hsu: It's insanely difficult to do this kind of work. I mean, the amount of paperwork we have to go through to get access to, to data and, and it, it's really a, a significant overhead. It's a pain in the neck.
Alex Young: Yeah, yeah. I mean, if you want to maintain access to a bunch of different datasets, you almost need a sort of administrative assistant to work almost full time on maintaining all of the facts and yeah, that It's, it's also, you know, the, the regulations on what you're allowed to do. One of the methods I developed that's been quite influential is a method for, you know, filling in missing data in, in in families. So like missing, missing genotypes of parents in a family. And there's actually been a move people seem to almost wanna create problems where there really aren't any. There's been, like, a movement, this publication in, in the BGA journal or something saying that doing a simple deduction about Mendelian inheritance in a family is actually unethical and potentially illegal.

So, so just to give you an example, they're claiming that if you say you have a, a parent that is homozygous. Say they have the AA allele at some position. You have an offspring that's heterozygous, say they have the AT allele. Then barring very rare de novo mutation, you can infer that the missing parent, you don't have the genotype 'cause the T allele.

That is apparently illegal and ethically problematic according to some segment of the human genetics world now.

Steve Hsu: I won't even mention what my company, Authom, does to solve crimes genealogically.

Alex Young: Right, yeah.

Steve Hsu: Yeah, better not to go there. But yeah, it's crazy. I you know, I sort of felt, you know, that we had hit peak wokeness and we were recovering, but maybe the ratchet just can't be turned back on a lot of this stuff. So I, I don't know.

Alex Young: But I think, I think it's also that people, people see this kind of stuff as a way to advance their careers, you know? They find something that they claim is, like, an ethical or a legal problem. Yeah. And they get a paper out of it. Yeah. And they advance their career. And then they get yeah present themselves as being really ethically switched on and, and get the grants for that kind of stuff. So there's kind of a bureaucratic inertia with, with this kind of stuff as well that goes beyond the sort of broader cultural milieu Yeah
it exists within.

Steve Hsu: Yeah. Absolutely. I mean, if you have a class of people whose job it is to point out bioethical conundrums by writing a paper which talks about a bioethical conundrum and then causing it to become a conundrum, a problem, then that, that's, that's how they advance in their careers, right? So

Alex Young: Yeah, I mean it's unfortunate similar with some of the, some of the embryo stuff as well. It kind of creates a lot of work for, for bioethicists saying that that's bad somehow.

Steve Hsu: So I think I've had you on for a while now.

Steve Hsu: one of the topics we had on our list was to talk about your cancer. Andjust curious how it's- I mean, you, you should just say whatever you wanna say, but don't, don't, you know, don't feel like you have to disclose anything that you're not comfortable disclosing. But I understand you've been fighting this for some time. It's quite serious. It's probably had some effects on how you view your work and your scientific legacy and life in general. Just I'd love to hear anything that you wanna share with us.

Alex Young: Yeah. Thanks. Yeah, I mean, it's, it's been, it's been a tough time. It was yeah, two just over two years ago, at the age of 35, I, I had zero risk factors, zero family history. I just had some digestive symptoms, you know. I had a colonoscopy and they found, you know, a large tumor.

Unfortunately two years later I still haven't been cured. So I've been through a lot already. You know, the first year I had like six months of chemotherapy and radiation, then I had four different surgical procedures and it all went well. I had a good response and it looked like I was cured actually.

And this is actually another area where it's funny, I didn't know so much about it as a scientist, but as a patient, seeing the way genomics really does seem to be actually revolutionizing cancer. And one of the ways it's doing it is actually through circulating tumor DNA monitoring, which has become a very important part of my life.

So actually Natera, who also do some of the PGTA generate some PGTA data, they they have a, an assay called Signatera which, which detects tiny amounts of DNA originating from the tumor in your blood. So that, that is a highly prognostic marker for rec- disease recurrence, so that it's becoming standard of care now that, that, if you have cancer you go through, you know, some sort of surgical resection, whatever, then they will often start monitoring.

So they look at the tumor genome and they, and they compare it to the normal genome and they put together like a panel of mutations. There's, there's a new company now called Personalis doing an even more sensitive test based on whole genome data whereas Natera just used the exome. And it's actually very, very sensitive.

It can detect tiny, tiny amounts of, of cancer before you can detect anything on a scan. So that was negative after all, all I went through in the first year. So it actually looked like I was probably cured. Like I think, you know, the probability was maybe like 10% of recurrence so I thought I was done. But then six months later the the circulating tumor DNA test turned positive which is very bad new was, was very bad news.

And then, yeah, they found something in a lymph node. I, I'd had a metastasis already in my liver, so I guess that put me at high risk of recurrence anyway. So they actually removed a third of my liver in the first year and and then, yeah, they found something in a lymph node of my liver. I got that had more chemotherapy for a couple of months, then they treated it with radiation. But then the ctDNA didn't go back to zero after that, and I went back into treatment. And yeah, then I had this other one thing about going through something like this is that it'll reveal anything weak in your mind or body, even things you didn't know about. And I guess one issue I had was that I had a H. pylori infection, which put me at risk of a duodenal ulcer. So I ended up nearly bleeding to death a couple of months ago. The cancer seemed to be under control, then I nearly bled to death from this ulcer and ended up having to have, like, an emergency procedure and whatever, and then that seems to have, like, revealed more cancer than we thought was there.

So now there's, like, three new disease sites in my abdomen and which can be treated with radiation. It's not, like, clear if that will cure me. There's also, like, experimental immunotherapy drugs, treatments that could be curative for me but accessing them is difficult. That's one thing I've learned from this process is it's a lot harder to access kind of investigational medicines than I feel like it should be.

So yeah, it's not, not, not clear what's gonna happen, and it does it does change your perspective on life a little bit. I mean, I kind of yeah, I kind of see, see my future very differently now. Like, I don't know how much of a future I'm, I'm gonna have, to be honest. And yeah, made me less afraid of other things in my life, like getting canceled.

Steve Hsu: Well, one of the things people always say hypothetically is, like, "Oh, if you only had X period to live," suddenly that might sharpen your perception of what's important in life and what's not important. Are there any do you have any insights in that direction?

Alex Young: It's definitely something people say. I think I've had some flashes of that, but it's one of those situations where it's not like I've been given, like, a, a terminal, a definitely terminal diagnosis. So I'm kind of existing in this sort of superposition of states where, like, maybe I'll get cured and maybe I won't, and it's, it's actually very hard to take any actions or, or make any plans and that.

It's made me just a bit more, a bit more nihilistic I hate to say, which is maybe not what you're meant to say. But I, I struggle to, to care as much about things I used to care about, which maybe means they weren't as important as I thought. And I just have to try and, like, live in the moment a bit more and be like, "Oh, whatever, I'll see what happens."

Steve Hsu: Obviously I'm not in your situation, but I am older than you and one of the things that happens with me is that I look back at my, say, my career as a theoretical physicist and I there are things that I cared a lot about when I was your age that now that I'm 60, or getting close to 60 you know, they just don't matter anymore.

You know? It's like, oh- Mm-hmm I, I really, really cared that, you know, I did get cited or I didn't get cited for this work that I did and, and now it's just like right who fucking, fucking cares? Like, it's like, it's like, wow, I got bent out of shape over that. You know, like, or, or whatever. I, so, I, I don't know, I do feel my perception of, of what is important and what's not important has changed over time.

Alex Young: Yeah. Yeah, I definitely care a bit less about the sort of academic career games that everyone Yeah is so focused on in academia and it's like, okay, I mean, I guess maybe it's because I'm not sure how much of a career I'm gonna have that I don't care about them. But it kind of makes you realize that a lot of those things you're doing, it's really just motivated by career progression.

Steve Hsu: Yes.

Alex Young: And it's not motivated by do I actually wanna do something that matters, you know, that has I think I maybe am valuing a bit more, like, things that have a real world impact. You know? I see it's actually interesting going, I've learned quite a bit about oncology and it's definitely a more, like, pragmatic field than human genetics.

And it's more focused on, like, improving real, real world outcomes. It has its issues too, for sure, the way that things run in trials and the thinking of oncologists I don't always agree with. But, but it feels like a bit more of a grounded feel to me often than, than human genetics. And then it does sort of make you question, like, oh, how much is any of the stuff that people are doing, like, really, really helping people, you know?

And, and actually it kind of made me a bit more a bit more keen on all the, the embryo selection stuff, 'cause that's something I, I feel like that, that can have, like, a real positive impact on the world in a way that a lot of academic research doesn't.

Steve Hsu: Yeah. Well, obviously I, I agree with you as an embryo selection company founder. Yeah. I worked on very esoteric stuff in theoretical physics like, oh, you know Yeah can you stabilize a wormhole with a cosmic string or s- you know. And on the one hand, I, I might look back and say I don't regret working on that because these are very fundamental objects in the universe and, and so Yeah learning about that's okay. But writing some little paper relevant to some experimental result or something that, like, everybody's forgotten after 20 years, of course, like, I just think, like, "What was I thinking? What a waste of time to work on that," you know, at, at the time. So yeah, so it definitely adjusts my perspective.

Alex Young: Yeah. Yeah, I think it makes you realize that the sort of career incentives in academia are, are quite misaligned really. It's, it's

Steve Hsu: Yes.

Alex Young: Yes it becomes a weird sort of insider's game really about getting grants and stuff, and I, I've actually been pretty I managed to get two, two i've only applied for two R01s and I got both of them, so I've sort of been very successful at that.

But, like, it's also and it's the next, you know, you have to get funding from somewhere, right? But it sometimes feels like that takes over really. Like, the winning funding becomes the thing everyone's optimizing for rather than, like, what are we actually making progress on important problems that might have a real world impact, you know? That sort of becomes the secondary concern.

Steve Hsu: Yeah.

Alex Young: Or people don't even people don't want, in human genetics, a lot of people don't want that much of a real world impact. Or maybe they're okay with certain, like, you know, maybe if you use a PGS to decide who gets screened for prostate cancer, that's okay, or, or, or to discover a drug target.
But yeah, I mean, I guess you can, you can argue that some of these kind of questions about, like, group differences, selection, that kind of only of academic interest. I'm not sure that's totally true either. But I still very much value truth after all of this. I think, I think finding the truth about societally important topics is also because, you know, these, these questions are important, and if we mislead ourselves about them, then we're gonna go astray. So yeah, it's, it's maybe radicalized me a bit more on why don't people actually answer the important questions, you know? Yeah. Both in terms of practical impact and societal importance.

Steve Hsu: Yeah. Absolutely.

Alex Young: And I think the main, the main reason is people are afraid. People are afraid of doing it a lot of the time. I mean, it's, it's also we don't have the data for some of them, but that's also 'cause people have, people have not prioritized getting the right kind of data, or they've explicitly blocked it as well.
Steve Hsu: Yeah. On, on the group difference issue though, I will say there's a, maybe not a principle, but a coherent perspective that some people will give me which is the following. Steve, why do you wanna purs- Not that I worked on this very much, but, you know, they would say, "Why would one want to pursue this?"
Because suppose you learned the answer are you really gonna be able to help anyone, or are you just gonna create more societal problems if this gets out? That, you know Right there is group differences. And of course, that was before we had the technology where we, we actually could give people who don't have who have, say, lower polygenic scores on a particular trait, we could give them more resources to have more embryos to select from, and actually we could transfer genetic luck from, you know, the population that has it alread- has already the, the right, the, the better sort of polygenic scores to the people who have lower polygenic scores.

So now Yeah actually have systems where you could help people. But prior to that, I think a lot of people could give a reasoned argument saying, "Yeah, I don't think you wanna look at that 'cause there's nothing we can do about it, and it's just gonna hurt people if you, if, if the truth comes out a certain way." But, but now it's, now we have Right the technology to actually fix it.

Alex Young: But it's also, like, people seem to have assumed the answer before they've actually done the studies as well, for one. So that, that's always bothered me about that kind of, like, moralistic reasoning in that domain. It's like, well, you've kind of already assumed the answer's gonna come out in a way that is not you wouldn't Yeah think is, like, socially good or whatever.

Steve Hsu: They, they might just say, "Look, there's some chance it'll come out this way

Alex Young: Right

Steve Hsu: which would be very damaging to social cohesion, and, and by the way, we can't fix it, so just leave it alone." Right? So they don't have to make that strong assumption about where it's, the answer's gonna be.

Alex Young: Right. Yeah. I mean, I think, I, I think you're right that people maybe attach a positive, a more positive moral valence to environmental causes and a more negative one to genetic causes for various outcomes, especially socioeconomic ones, because they think that genetics is fixed and the social environment can be manipulated, which I actually think on both points people are wrong about. I think Yes.

Steve Hsu: Yes

Alex Young: manipulating sociocultural things to get the desired outcome doesn't really have a very good track record. I'm not saying it's, like, impossible, but it's not easy. And actually manipulating genetics in theory is far more tractable. You know, you have this finite molecule that we understand a lot about.
We can even edit it. We can, we can make predictions based on it and select them. I mean, so to me, I don't really see there being a big moral valence in whether something is genetically or environmentally caused, but a lot of people clearly do.

Steve Hsu: Yeah. But I think most of those people have not thought through how embryos, how well embryo selection could work or even someday embryo editing. And so they don't realize, like, we've switched into this era where we, we had no way to address genetic problems of this kind to, well, in the foreseeable future, we will be able to actually intervene and help large populations of people if they need help.

Alex Young: But I think a, a lot of those same people are probably against making genetic interventions, so.
Steve Hsu: Yeah.

Alex Young: But then you've just kind of like, you've decided everything in advance. You've decided you don't want to know what the cause is. Yes. And then you're not gonna accept any possible solution if the cause comes out one way versus the other.

Steve Hsu: I think it's completely incoherent given where we are now technologically. But maybe given where we were technologically 50 years ago, it wasn't, there was a kind of coherent case they could make that just leave this alone because we're not gonna be able. Right. We can't currently do anything about it.
Alex Young: Yeah. I mean, I understand why it's a sensitive topic, but, you know, in the long run, the data is gonna come out one way or the other, whether people like it or not.

Steve Hsu: Yes. I agree. But these people are running out the clock as much as they can.
Alex Young: Yeah. Maybe. Maybe that's part of it. People, people think it can, it, it's better off being. I mean, maybe it has been better off being delayed until we have the right kind of data to answer it rigorously. I, I, I would maybe agree with that.

Like, you know, this mix, this within-family admixture design, I think is, is pretty powerful for answering those topics. And if you look at what the you know, the recent selection paper from, from David Reich's group, they. Yeah, I mean, maybe not every result in that paper is gonna hold up and be completely robust.
But I think that, you know, you're, you're seeing in that paper the, the, the scale of ancient DNA and the power of polygenic scores, especially if we could get one stride from family-based studies that, that have less potential stratification issues. We can see the power there to, to, to understand selection forces as well and Yeah how they might have been the same or different in different populations. So, you know, I, I think that it's not gonna be tenable to not look at these, these questions going forward. And I think you're seeing, you know, David Reich and, and Peter Visscher are two of the biggest names in the field, and they're, they're kind of breaking these taboos a little bit right now. So I think that we are, we are gonna enter a, a different world in terms of, you know, real scientists studying this with powerful data in the near future. I
Steve Hsu: think for, again, for the listeners who are not expert on this, so Alex is talking about a paper that was circulating for, I think, years, and finally got published from the Reich group. And as mentioned, Reich is one of the leading figures especially with ancient DNA. And this is evidence that on timescales of thousands of years you had sort of, of order one standard deviation changes in some polygenic traits due to what appears to be selection. And the, the, the size of selection effects might be different in different regions of the world or different civilizations, et cetera, et cetera.
And all of this now can be studied in a more quantitative way.

Alex Young: a, There's a new preprint they just released looking at the same, using the same methods basically in ancient East Asian genomes. And they actually see highly correlated selection signals in East and West Eurasia, which is, which is pretty interesting.

Steve Hsu: Yeah. That's a super interesting finding.

Alex Young: Yeah

Steve Hsu: yeah. I mean, all these things have been theorized a lot, but now we have the data to talk about it. Like, how did farming affect, you know, what, which traits were selected for when you switched from being a hunter-gatherer to being a farmer? What happened when you started living in cities? I mean, all these things are super interesting.

Alex Young: Yeah. Well, I think that the idea that selection has not been important in the past 10,000 years, or even the past 100 years, I think it's not really I mean, it's, it's a bad null model really. Like, you're always gonna have some kind of selection. Like, I was involved in this paper while I was at DeCode where we saw that the, the genetic propensity to education was actually declining over the 20th century in Iceland.
And I mean, you know, every time you look at the data, you see that there's, there's always bound to be the, the correlation between fertility and some complex heritable trait, it's never gonna be zero. It might not be strong, but so the question is not, is there selection or not? The question is, how strong is it and in, and in what direction?

And you know, that Akbari Reich paper, interest- One of the interesting things I saw in that was it seemed, you know, if you believe that education and intelligence selection results, and, and I think that, that they seem to be reasonably robust from triangulating different, different sources of data.
But interestingly enough, it looked like the selection was strongest in the Bronze and Iron Age, and then basically stopped in the, in the classical period. So yeah, you wonder what was going on then to generate such strong selection on those, those education and intelligence associated variants, and why it would then stop in the sort of classical period. But that seems to be the signal they found.

Steve Hsu: When I look at their graphs, sometimes I see like three dots, right? And then some interpolating band. So I, I, I don't know how strong

Alex Young: They've, they've got a, they've got a finer grained plot in the supplement actually- Okay ... for different periods. And you can see the selection strength really goes down to zero ar- around like 0 AD or something for IQ and education. So I don't know if, how much you should, faith you should put in those plots, but

Steve Hsu: Yeah

Alex Young: it's, it was interesting nonetheless. Yeah. I guess it makes sense to me that farming would I mean, this is the Greg Cochran, Henry Harpending argument, right?

That agriculture created a very different environment, and that would've selected for different, you know immune all sorts of different things would be sl- selected for in that drastically different environment to being a hunter-gatherers. I mean, that, that argument always made sense to me, and I, I guess now we're sort of seeing the, the signal from that

Steve Hsu: I would've been surprised if it were otherwise. I, I, I would, I would've expected pretty big changes in what was being selected for.

Alex Young: Yeah.

Steve Hsu: Yeah. Well, Alex, I, I appreciate you spending time with us, and always fascinating to chat with you. I, I wish you all the best in fighting your cancer.

I, I wonder if you're interested at all in getting involved in trying to figure out how to do this monitoring of tumor cell presence, circulating tumor cell presence more accurately. That seems like an area that your skills might actually could be applied to.

Alex Young: Yeah, you know, some, some of the algorithms are probably not that dissimilar from, from using this low coverage Yeah pGTA data. But, yeah, no, I, I looked into it a little bit and I, I mean, I think, obviously moving from the exome data to the whole genome is kind of like an obvious thing, and there's a company already doing that. It's not... I thought about maybe using like long, you could use like long read sequencing- Mm-hmm to to improve the detection of tumor derived fragments. But apparently the cell-free DNA that ends up in the blood tends to be very short fragments anyway. So yeah, I'm sure there's more to be gained than what people are doing now, but it hasn't, it's not obvious to me how you'd improve it. I mean, it's already incredibly sensitive actually.

Like, they've done a pretty good job with it. Like, the the Personalis the Personalis test can detect one tumor one tumor fragment in part per million, basically.

Steve Hsu: Is, sorry to get into science on this 'cause I was just about to let you go, but- Is the tumor cell genome very different from a regular cell or is it just a few driver mutations? Is it ... I mean, how, how different is the, the genomes?

Alex Young: It, it, it very much depends on the cancer actually. So, you know, some early onset colorectal cancer cases, not mine, they're caused by germline mutations and mismatch repair genes. That's what's called Lynch syndrome. Mm. And those tumors are characterized by very high mutational burden and microsatellite instability.

So they actually have drastically different genomes and that actually makes immunotherapy interesting. Immunotherapy works better because the more mutated the tumor is- Mm ... the more you get these foreign antigens basically presented on the tumor cell surface for the immune system to recognize. Now, in other cancers that are

And, and, and this can happen if you have, for example, BRCA. Like BRCA mutated cancers often have very high mutation rates because if you, if the cancer originates partly from a driver mutation in a double strand break repair gene, some sort of DNA repair gene, then often the tumors will have drastically different genomes.

But for example, mine, I think the estimate was only 4.2 mutations per megabase, which is- Wow ... medium level mutation for colorectal. that's still enough I mean, every, every cell line kinda has characteristic somatic mutations, but cancer cells tend to have a higher mutation rate than, than non-cancer cells.
So there's usually plenty of mutations to use for the detection, but it doesn't always work perfectly actually. Like I think sometimes, they create a panel of mutations to detect the recurrence and it just doesn't really work. But

Steve Hsu: Mm

Alex Young: seems to work well for my tumor at least. Thankfully. Wow. It does give you an ear- it gives you an early warning and helps in managing treatment. Yeah. 'Cause you can see in the ctDNA, like if you start a new treatment you can see the response immediately in 'cause it's a quantitative measure of disease burden in, in, in many ways. So, so if you start a treatment you can see the level of tumor DNA in your blood go down like immediately, like the week after.

Yeah. So that tells you the treatment's working. So I don't think they've really fully integrated this into oncology management yet because, you know, before you'd have to wait three months to do another scan to tell if the treatment is working and maybe the disease could have progressed by then. But this is actually one of the issues I have with maybe the way they do things in oncology is they kinda want you to just keep going with like some treatment for ages until something bad happens and then switch treatment.

But really to me it would seem to make sense to like try something, see how good the response is. If it's not that good, try something else. See if the a and, and use the ctDNA to get re- almost real time feedback in how, how effective a treatment is working and you know, I think that GitLab guy who went Vander node on his cancer.

I don't know if you saw that story. He's kind of following. Yes. That approach and- Yes ... some of the more forward-thinking oncologists, like some of the oncologists I'm working with at UCLA and City of Hope, they, they're kind of doing something more like that. But I think it's, it's still kind of cutting edge a bit, really.

Steve Hsu: Great. Wish you all the best, and hope we can have another interview at some point in the future.

Alex Young: Thank you very much, Steve. Always a pleasure to talk to you.

Steve Hsu: Yep. Take care.