UCLA genetics professor Alex Young on polygenic embryo selection: how IVF can halve disease risk and why academia punishes researchers who get involved

Speaker 2: amplifying and sort of playing into it and hopefully we can get Adam back on the show and ask him more about it. Anyway, we have our next guest live here in person in the TV and Ultra Day. Let's bring in Alex Young. He's a professor at UCLA. Welcome to the show. Hey. How are you doing? Good.

Speaker 7: Yeah. Nice to meet you. Thanks so much for taking the time to come down to the show in the studio. Why don't you introduce yourself first? Yeah, I'm Alex Yang. I'm a professor in statistical genetics at UCLA med school and I'm also an advisor to Heracyte, a company doing advanced genetic testing in IVF. Yeah. What drew you to Heracyte in particular? Well, I had a realization that this technology, so the technology that they've developed is applying genetic testing to embryos so you can predict their disease risks or traits. I had a realization that that was a very powerful technology and an incredibly important application of a lot of the research that I was doing. But there were already some companies in that space and I didn't find any of them that serious about doing really innovative R and D. And Heracide actually approached me in 2023 with really an intriguing idea to develop an algorithm and I kind of couldn't resist doing that. So I decided to get involved and ended up, yeah, creating some good technology.

Speaker 2: What have been the key milestones in embryo selection technology? Like how much of it is machines in the lab better data collection versus more on the algorithm computational side?

Speaker 7: Well, I mean, has been around for a while. Yeah. Genetic testing in IVF is not new. Yeah. So, you know, I think even back in the '90s they were doing sex selection to avoid certain sex linked diseases.

Speaker 2: And

Speaker 7: there are other tests already developed such as PGTM, which looks for like Mendelian disease genes. Yeah. So they're like cystic fibrosis or something like And then also tests to look for chromosomal abnormalities PGTA. So those tests were more about being able to get lab a lot of that was lab innovation, being able to get the genetic data from the embryos. Now, this new wave of technology that Heracide and some of the other companies are developing, that's grown out of the research program in human genetics, which is much more computational. It's been enabled by the massive drop in the cost of sequencing technology that's happened over the past twenty years. That's created huge genetic data sets and these big, what we call bio banks places at UK Biobank was like the flagship one where you have like 500,000 people Sure. With their whole genome data and all of their medical records and things like education as well. That's enabled us to create these predictors called polygenic scores. Sure. And they're the things that we can now apply in IVF and that's sort of more of a computational aspect than a lab aspect, I would say. Yeah. Yeah.

Speaker 2: Give me some history on the UK Biobank. It feels like big step to share that much information publicly. I assume it's somewhat anonymized, but what are the considerations with sharing genetic data and associated traits? Obviously there's a medical benefit, but how do individuals grapple with that trade off?

Speaker 7: Yeah. Well, U. K. Biobank kind of took a much more liberal attitude towards allowing people to access that data. So, you know, it's kind of sad actually that so they were kind of the first big biobank and it wasn't originally designed purely for genetic data. It was designed for sort of medical epidemiology and then they added genetic data to it. And they made it very easy for both companies and for academic researchers access that data and it's had, you know, 100x the impact of other biobanks including, you know, The U. S. Has been quite behind actually in that respect and a lot of other genetic data sets are really hard to access and are actually you're actually barred from accessing them for commercial uses. So, yeah, I would commend UK Biobank for taking this more liberal approach that you have to protect the informed consent and privacy participants, of but I think sometimes the restrictions that are put on other data sets really have hampered the development of both the research side but then also the commercialization of ergonomics.

Speaker 2: How powerful is the technology these days? Like how good are we at detecting diseases? I imagine that there's some diseases that we're very good at. What's like the best case study right now? Yeah,

Speaker 7: so I mean, if you're talking about in the sort of IVF context, where that's most powerful is diseases where you have a particular gene, a particular position in the genome that has an outsized effect on the risk of that disease. So if you look at something like type one diabetes or Alzheimer's, there are certain genes in the genome that have a really large impact. And if you're someone if you're a couple and one of you carries one of those major risk genes, then doing this embryo selection can drastically reduce the risk. Can probably pretty much eliminate the risk of passing on type one diabetes, for example, drastically reduce the risk of passing on Alzheimer's disease. Now, there are other what we'd call more complex diseases like type two diabetes or various cancers and there it's a bit less effective, but it can still be quite effective. You can roughly say half the risk your offspring gets some disease and if that disease is present in your family history, then that can be quite a substantial absolute risk reduction. So you could go from something like twenty percent risk of type two diabetes to ten percent risk in your offspring. Yeah. What's the progress

Speaker 2: has progress been faster on the screening side or on the actual drug development and treatment side? I mean, we've seen the peptide boom, the GLP-one boom. There's a new major biotech acquisition every few days now. It feels like the pace is accelerating in actual drug development. But what are you seeing actually being unlocked by new technologies?

Speaker 7: Well, yeah, human genetic data has been incredibly important in drug development. You've seen big pharmaceutical companies like Regeneron. They hired a lot of the best people from I was at Oxford for my PhD and like a lot of the best people from Oxford ended up getting hired by Regeneron to set up like a big new genetics center. And they've amassed like a huge dataset there, like 10,000,000 people or something now. Yeah. And that's definitely driving like target discovery in the pharmaceutical industry. So that's been where a lot of the investments come. This sort of IVF and screening side has been maybe a more neglected part of the potential of human genetic data, partly I would say for political and ideological. There are lot of people who are opposed to this technology. Sure. So, you know, I suffered some career consequences for getting involved in

Speaker 2: this industry myself. Yeah. What happened?

Speaker 7: Well, you know, when it came out that I'd been involved in heresy, I mean, some people said some unpleasant things about me, but I don't really care about that. But I also had some collaborators pull out of a paper I had in review at Nature and I also had a job offer rescinded

Speaker 2: from a top US university. Oh, interesting. Yeah. So how are you splitting your time these days?

Speaker 7: Well, I'm mostly, you know, doing my academic research. But, yeah, my main contribution to Heracite was this algorithm that we developed and we're trying to get published now. So yeah, involvement is more as like an adviser now, but I did make a sort of key technological contribution.

Speaker 2: Yeah. What what does an algorithm like, are we talking about, like like, machine learning, transformer based AI? What what or just, you know, some statistical model? Like, what goes into honing a a, you know, an algorithm that actually moves the needle on predictability?

Speaker 7: For predictability, it's it's it's more kind of statistics and machine learning than AI. Like, it turns out that you don't need these very complicated kind of like nonlinear models that transformers and large language models are particularly adept at. So for example, creating this like Heracyte created this very powerful genetic predictor of IQ, much more powerful than anyone thought was possible and that was sort of they used some deep learning based models to kind of create better curated data, psychometrically curated data. And then when you're actually creating these predictors, this is basically it's kind of like a Bayesian statistical model where you're using functional data, so information about which genes are likely to affect the trait to improve your estimates. The algorithm that I developed wasn't really on the prediction side. This was more to lower the technological and regulatory barrier to accessing the technology and this was a machine learning model based on hidden Markov models. And this basically enabled it turned sort of a routine test. So there's a routine test for aneuploidy for chromosomal abnormalities such as Down syndrome. And this algorithm enables the data that's produced for that routine test to then be used to give a comprehensive genome profile of each embryo. And one advantage of that is that that test is done as routine in like sixty percent of IVF cycles in The U. S. Worldwide. It also kind of puts the power back in the hands of the couples undergoing IVF because of HIPAA legislation in The US and GDPR in Europe, you actually have a legal right to that data so you can just request that data from your clinic in whatever country you're in and send that to Heracite and then that enables you to get all these disease risk predictions and trait predictions like IQ from routine tests. There's kind of there's two different sides to where the algorithms come in. One is creating these predictors and the other is actually how do you get the genome data on the embryo because that's not a trivial technical problem either. Where does the FDA stand on this? Where do they fit in? How do they actually regulate tests?

Speaker 2: Walk me through all that. So

Speaker 7: in The US, the IVF space is pretty unregulated. Okay. I think it's because

Speaker 1: Is that good?

Speaker 7: Well, it's it's I I think it's on that good because if you look

Speaker 1: at the Like, I'm normally pretty against regulation, is a big decision. Feels like maybe an area that you would want more government involvement.

Speaker 7: Well, you know, I guess my take on this is that in an ideal world, you'd have a kind of light touch regulation that ensured some unscrupulous players were kept out of the market. Like I know you had Nucleus key in on here before and they've had some test results that looked pretty misleading to me. And you'd want to make sure that misleading test results are not being given out to people. But my concern with any kind of regulation is that it's going to suddenly evolve into something that just completely kills innovation. Like, this is basically impossible to do in Europe, for example. So all of the companies doing this are US based and that's partly because it's a pretty unregulated space. So it's enabled this industry to actually grow here, whereas it's not

Speaker 1: possible in other countries. Do you think the industry is ready for venture capital? Yeah. Yeah, I think so. I mean, I guess because I because and I ask that because it feels like there's a lot of promising work being done but unclear that any time you involve large amounts of venture capital, then then people have various growth expectations and decisions get made that maybe wouldn't be in the long term interests of patients.

Speaker 7: Yeah. I mean, think that's certainly a risk, but there is a lot of demand for this product and I think that we're still in the sort of infancy. And part of what's holding this industry back is we kind of need more we need more money to put into it. We need more awareness of the potential of this technology. And know, Heracite, revenue has been growing pretty well. They're trying to do a big raise this year. I think that there could be kind of a preference cascade here. Like a lot of the people that have been a lot of the customers so far have been people from the tech world, people that are quite wealthy. And I think that's going to create a lot of demand as as people see as people see these these wealthy and sort of elite people wanting these services, then I think everyone else is gonna want them. So I think I think it's it's poised for for a big growth spurt. Is one of the big challenges of the feedback

Speaker 1: cycles where you have models today that are making predictions about outcomes that won't happen, for decades? And, so it's hard to gauge act you know, basically accuracy of predictions if if we basically just need to wait and see.

Speaker 7: I mean, that that is a concern that that the predictions might, based on historical data, might not perform as well in the future. I mean, I think that is not a concern that's exclusive to this kind of technology though. Like a lot of what is done in medicine is based on randomized controlled trials that were decades ago I don't see it being hugely different here. One good thing about this technology is that rather than having to do this big long, I mean, to, it's sort of impossible in a way to, if you wanted to do a randomized controlled trial for some kind of late onset disease, you'd have to wait like sixty years for people to get cancer something. It's basically impossible to do that. But Nature has provided us with a natural experiment that recapitulates a lot of the properties of a randomized controlled trial. So this is related to my academic research in that so embryos are basically siblings genetically and if you can predict genetic differences between siblings, then you can be pretty sure that that's going to also translate into predicting difference between embryos. So it's a bit and that's kind of the technical standard that we've tried to establish at Herocyte that you need to validate all of your predictors carefully in predicting differences between siblings within family. That ensures that you're getting like a causal estimate of what these predictors are able to do. And yes, there might be some differences in the future like if the environment changes drastically, there's some different environmental effect that puts someone at risk of cancer. It might not capture that so well. But that's just the nature of the game in a way. We to use the information we have now to make the best predictions we can about the future. Yeah. What's been

Speaker 2: your experience with the health system broadly? I mean, you done IVF yourself? Have you dealt with advanced medicines or anything beyond the normal Advil after a long weekend?

Speaker 7: Yeah. Well, I am actually a cancer patient,

Speaker 2: which is I'm sorry.

Speaker 7: Yeah. So I've had a lot more dealings with the health system than I would like. Yeah. What have the learnings been? What's the process been like? One thing, I actually had to do fertility preservation Okay. Before undergoing radiation and chemotherapy. So that actually made me feel a lot more pro IVF. Sure. So basically I had to freeze my sperm and basically now I can only reproduce through IVF. Yeah. So dovetails a bit with the work I've been doing with Heracite. But yeah, I mean Have you tracking

Speaker 2: I mean, it seems like you're very capable of doing research on your own. Have you been doing research into your own cancer to understand it at a deeper level? Do you feel like there's more self education happening now?

Speaker 7: Do you have a unique position there? Yeah. Yeah, I have a bit. It's one of those things like I've seen these stories about people kind of using AI to sort of figure out some sort of mRNA vaccine for the cancer.

Speaker 2: What's your reaction to that?

Speaker 7: Well, you know, some I can't remember what the guy's called. We had him on the show with Yeah. Yeah. No, that was quite an inspirational story. And I've talked to some of the people that have been working with him recently about that. And yeah, I think the standard way that sort of oncology is approached, it doesn't seem to be as patient centric as it could be. It's not like, are we going to throw absolutely everything we can at curing this person? It's more like, oh, we're just going to sort of follow the standard playbook and wait until something really bad happens. So I think there's definitely potential for people to take more agency with serious conditions like cancer and using genomics and AI to enable that, although it's not trivial to do that well. I mean if if you're a billionaire, it helps. Mhmm. You can get all of these experiments done that are not standard. Yeah. But, yeah, actually doing it well, I don't think it's it's it's not as simple as just plugging your data into an LLM then you have a have a cure, you know? Yeah. Yeah. Yeah. What's the bigger bottleneck?

Speaker 2: Like more larger datasets that are publicly available for like the new like the English biobank? Or is it more computational or just general intelligence?

Speaker 7: Well, think in terms of cancer, it's probably more getting more making clinical trials easier. Okay. There's this movement online and there's clinical trials abundance movement. Sure. It does seem to me that, like in The US especially, clinical trials are very expensive. Like there's a treatment that I am trying to access that doing the phase three trials for example in Australia instead of The US because it's too expensive. Yeah. Then there's some and then trying to get into a trial is like threading a needle, you know? So I think that making it easier for patients to access cutting edge treatments and

Speaker 2: just And then even if you get in, it would be placebo controlled so there's a fifty percent chance that you don't get Is that right? Yeah. It's one of those things where wearing

Speaker 7: my scientist hat, I'm like, oh, it's really important to do placebo in a properly controlled trial so you can make the right inferences. But then when you're you're in a patient, you're like, I don't want to be in placebo. I'm like, what's the point in that? Yeah, exactly. Yeah. Yeah. We heard

Speaker 1: about this with GLP-one trials. Yeah. Where people would the drug's so effective, people would realize they're in the placebo and then just drop out of the trial entirely.

Speaker 7: Right, yeah. Mean I was thinking that about this trial. Was like, well, if it it's probably going to be pretty obvious within the placebo because I don't have any like side effects and if it's just not working then why not just drop out? But they kind of like expect you to just keep on going in these trials when things aren't working which also doesn't seem to be particularly in the patient's interest. Yeah.

Speaker 2: Well, what's next for you? What do you want to focus on?

Speaker 7: Well, you know, getting cured of cancer would be nice. Yeah. Seems like priority number one. Yeah. I'm I'm, you know, I'm pretty excited about this whole reproductive tech space. Sure. I think there's a lot of potential. It's it's we're really in the infancy this of this industry. So what Heracide's been doing with the polygenic scores, they've created very powerful predictors with that. But these polygenic scores, they only capture a certain fraction of the total effect of the genome on disease risk and traits. So actually a lot of the remaining signal that we could get out of the genetics is in these rare damaging mutations that are not typically included in these scores. And that's one of the things that Herocyte's been working on. There's an important application of AI models like AlphaFold, AlphaMissense

Speaker 2: Sure.

Speaker 7: And sort of DNA foundation models to try and predict what these rare mutation, what risks these rare mutations might confer. Interesting. And including, know, Heracide have already developed a product looking at how you can predict risk of neurodevelopmental disorders using these kind of rare variants. I that's going to be an important area going forward. Yeah. Then, you know, this more like further in the future kind of technologies, in vitro gametogenesis, so that's a technology to create gametes, so sperm and eggs. Eggs is what would be more useful generally from adult cells, not So gamete that could potentially create like thousands of embryos. So I think you could, you could, and then there's also gene editing actually. So can use CRISPR or some technology like that to go into an embryo and edit particular base pairs, remove some disease causing variant or maybe more controversially enhance some ability like you could put in the Tibetan altitude adaptation gene that Nims Dai has or you could put in a sprinter gene or something. Sure, sure. So I think my idea for the future of this space could be that you have this kind of stack where you have in vitro immunogenesis that creates thousands of embryos, then you can get the genome data, do the predictions of disease risks and traits from that, select a few promising embryos and then do some edits in them. Yeah. So that kind of

Speaker 2: excites me that I think. Sci fi. Yeah. Well, thank you so much for coming on and We sharing it with