There’s polygenic screening now. It doesn’t include eg IQ, but polygenic screening for IQ is unlikely to be very good any time in the near future. Probably polygenic screening for other things will improve at some rate, but regardless of how long you wait, it could always improve more if you wait longer, so there will never be a “right time”.
Even in the very unlikely scenario where your decision about child-rearing should depend on something about polygenic screening, I say do it now.
Polygenic predictors have improved since Gwern’s 2016 post on embryo selection. Using his R code for estimating gain given variance and standard deviation and taking the variance explained from the Educational Attainment 3 study, I find that selecting from 10 embryos would produce a gain of between 4 to 5 points for the top-scoring embryo (assuming no implantation loss). Accounting for implantation loss it would probably take 14 embryos or so to get the same benefit.
Steve Hsu thinks that if we were to offer UK biobank’s IQ test to a million participants, we could get IQ predictors that would explain 50-60% 30-40% of variance. That would work out to a gain of 9-10 IQ points from selecting among 10 embryos, and up to 14 points if you had about 30 to choose from. See “technical note” in this post: https://infoproc.blogspot.com/2021/09/kathryn-paige-harden-profile-in-new.html
Embryos produced by the same couple won’t vary in IQ too much, and we only understand some of the variation in IQ, so we’re trying to predict small differences without being able to see what’s going on too clearly. Gwern predicts that if you had ten embryos to choose from, understood the SNP portion of IQ genetics perfectly, and picked the highest-IQ without selecting on any other factor, you could gain ~9 IQ points over natural conception.
Given our current understanding of IQ genetics, keeping the other two factors the same, you can gain ~3 points. But the vast majority of couples won’t get 10 embryos, and you may want to select for things other than IQ (eg not having deadly diseases). So in reality it’ll be less than that.
The only thing here that will get better in the future is our understanding of IQ genetics, but it doesn’t seem to be moving forward especially quickly, at some point we’ll exhaust the low- and medium- hanging fruits, and even if we do a great job there the gains will max out at somewhere less than 9 points.
Also, this is assuming someone decides to make polygenic screening for IQ available at some point, or someone puts in the work to make it easy for the average person to do despite being not officially available.
I am not an expert in this and would defer to Gwern or anyone who knows more.
Yes, generating lots of embryos will help, but the marginal returns decline extremely quickly because you’re taking the maximum of a Gaussian distribution.
Gain is proportional to sqrt(ln(number of embryos))
Going from 30 embryos (roughly the upper bound of what you can get with normal IVF) to 1000 will only increase gain 43%.
43% is still pretty big, but it’s not going to radically change the world. For that you’re going to need massive parallel gene editing, iterated embryo selection or whole genome synthesis.
I did a calculation a while back and estimated it would cost about $200 million to synthesize an entire human genome at today’s prices. But there’s a bunch of other technical challenges you’d have to overcome.
Gene editing seems to be improving but there are still some weird issues with CRISPR where it seems to randomly chop off chromosomes sometimes.
There are a couple groups working towards in-vitro oogenesis, which is probably the most important step to making iterated embryo selection work. I think the main blocking step there is making the environment in which the eggs mature. You need a bunch of follicle cells and the only source for those currently is abortion tissue. There’s a couple groups trying to make them from stem cells instead: https://www.technologyreview.com/2021/10/28/1038172/conception-eggs-reproduction-vitro-gametogenesis/
So lots of stuff in the pipeline, but nothing seems to be imminent.
(In particular, trying to understand what you’re saying made me understand something basic about IES vs just embryo selection: with IES you’re actually doing many iterations of sexual reproduction with a population, so you can get a genotype composed of any alleles that are present in any of the starting embryos.)
If you read the Technology Review article I posted, there is a Japanese team that managed to do it in mice, but it took them 4 years. I don’t really know that much about the technical details, but just judging from the empirical results it does not appear to be an easy problem.
I guess the social/logistic one is not wanting to create a solution that relies on a supply of abortion tissue, which leads to the scientific/technological one of how to create the follicle cells.
There’s polygenic screening now. It doesn’t include eg IQ, but polygenic screening for IQ is unlikely to be very good any time in the near future. Probably polygenic screening for other things will improve at some rate, but regardless of how long you wait, it could always improve more if you wait longer, so there will never be a “right time”.
Even in the very unlikely scenario where your decision about child-rearing should depend on something about polygenic screening, I say do it now.
Polygenic predictors have improved since Gwern’s 2016 post on embryo selection. Using his R code for estimating gain given variance and standard deviation and taking the variance explained from the Educational Attainment 3 study, I find that selecting from 10 embryos would produce a gain of between 4 to 5 points for the top-scoring embryo (assuming no implantation loss). Accounting for implantation loss it would probably take 14 embryos or so to get the same benefit.
Gwern’s code: https://www.gwern.net/Embryo-selection#benefit
EA3 study: https://sci-hubtw.hkvisa.net/10.1038/s41588-018-0147-3
Steve Hsu thinks that if we were to offer UK biobank’s IQ test to a million participants, we could get IQ predictors that would explain
50-60%30-40% of variance. That would work out to a gain of 9-10 IQ points from selecting among 10 embryos, and up to 14 points if you had about 30 to choose from. See “technical note” in this post: https://infoproc.blogspot.com/2021/09/kathryn-paige-harden-profile-in-new.htmlWhy not?
Embryos produced by the same couple won’t vary in IQ too much, and we only understand some of the variation in IQ, so we’re trying to predict small differences without being able to see what’s going on too clearly. Gwern predicts that if you had ten embryos to choose from, understood the SNP portion of IQ genetics perfectly, and picked the highest-IQ without selecting on any other factor, you could gain ~9 IQ points over natural conception.
Given our current understanding of IQ genetics, keeping the other two factors the same, you can gain ~3 points. But the vast majority of couples won’t get 10 embryos, and you may want to select for things other than IQ (eg not having deadly diseases). So in reality it’ll be less than that.
The only thing here that will get better in the future is our understanding of IQ genetics, but it doesn’t seem to be moving forward especially quickly, at some point we’ll exhaust the low- and medium- hanging fruits, and even if we do a great job there the gains will max out at somewhere less than 9 points.
Also, this is assuming someone decides to make polygenic screening for IQ available at some point, or someone puts in the work to make it easy for the average person to do despite being not officially available.
I am not an expert in this and would defer to Gwern or anyone who knows more.
Thanks!
And, hypothetically, generating lots of embryos to choose from? Or is that not in the cards?
Yes, generating lots of embryos will help, but the marginal returns decline extremely quickly because you’re taking the maximum of a Gaussian distribution.
Gain is proportional to sqrt(ln(number of embryos))
Going from 30 embryos (roughly the upper bound of what you can get with normal IVF) to 1000 will only increase gain 43%.
43% is still pretty big, but it’s not going to radically change the world. For that you’re going to need massive parallel gene editing, iterated embryo selection or whole genome synthesis.
I did a calculation a while back and estimated it would cost about $200 million to synthesize an entire human genome at today’s prices. But there’s a bunch of other technical challenges you’d have to overcome.
Gene editing seems to be improving but there are still some weird issues with CRISPR where it seems to randomly chop off chromosomes sometimes.
There are a couple groups working towards in-vitro oogenesis, which is probably the most important step to making iterated embryo selection work. I think the main blocking step there is making the environment in which the eggs mature. You need a bunch of follicle cells and the only source for those currently is abortion tissue. There’s a couple groups trying to make them from stem cells instead: https://www.technologyreview.com/2021/10/28/1038172/conception-eggs-reproduction-vitro-gametogenesis/
So lots of stuff in the pipeline, but nothing seems to be imminent.
Thanks, this is an enlightening summary!
(In particular, trying to understand what you’re saying made me understand something basic about IES vs just embryo selection: with IES you’re actually doing many iterations of sexual reproduction with a population, so you can get a genotype composed of any alleles that are present in any of the starting embryos.)
Hmm, that seems like it shouldnt be that hard of a problem to solve, but idk. I hope someone takes this on if that’s really a bottleneck.
If you read the Technology Review article I posted, there is a Japanese team that managed to do it in mice, but it took them 4 years. I don’t really know that much about the technical details, but just judging from the empirical results it does not appear to be an easy problem.
Oh, I assumed the problem was a social / logistic one, but now I’m assuming there’s also a scientific / technological one
I guess the social/logistic one is not wanting to create a solution that relies on a supply of abortion tissue, which leads to the scientific/technological one of how to create the follicle cells.
For a normal trait, the variance of the children of a fixed couple is approximately the population variance. I think that’s a lot.