I have a friend who’s a geneticist and is much more pessimistic about the expected impact of embryo selection on a given desired trait. I chatted with him and read some papers he suggested and I now have a lower estimate of the expected value of embryo selection than before.
Very curious to hear from other experts in this field who disagree, especially those who don’t have a financial conflict of interest!
It seems it’s quite unclear how to even measure how heritable a given trait is, with different methods giving wildly different answers. And even if a trait is highly heritable:
The variation between siblings (or rather embryos from the same parents) is likely to be quite a lot lower than the variation seen in the general population.
The variation that’s due to very rare genetic variants is likely not captured in today’s polygenic risk scores.
Some relevant papers, which I don’t claim to fully understand, but from which I’ve extracted the above as “the gist”:
“Therefore, the so-called SNV-based heritability gives the upper limit of the variance between people in their liability to a disease that can be explained by PRS and represents the variance explained by common DNA variants. As GWAS sample sizes increase, the variance explained by PRS will also increase and approach the SNV-based heritability. The SNV-based heritability estimates vary across diseases, but an approximate upper limit is approximately 30%. Although in principle, use of whole-genome sequence data could increase the variance explained by PRS (because more variance would be tagged by measured markers, ie, the SNV-based heritability approaches the heritability), it is unlikely (at least in the short term) to improve PRS (eAppendix in the Supplement). Risk stratification based on current and future PRS is illustrated in Figure 4.”
The supplement is mostly a verbal argument along the lines of “there are just huge number of rare variants, each of which exists in only a few people, so it will be difficult to figure out how to use them to improve phenotype prediction.”
Yes, I hear criticisms like those made by your friend all the time. They aren’t particularly discouraging to me.
The variation is not that much lower. The standard deviation of any continuous trait is about 70% that of the general population. That’s still plenty for selection.
The impact from rare genetic variants is mostly NOT captured in today’s genetic predictors, and that’s a place where the field could improve more in the future. But you don’t NEED rare genetic variants for embryo selection to work. You just need a predictor that correlates strongly enough with the actual trait for selection. And for many traits we already have that.
The paper you linked showing heritability is lower using relatedness disequilibrium regression was interesting. They don’t include intelligence or most of the disease predictors on which embryos are currently selected in the table, so it’s hard to give a good estimate of how strongly we should expect it to affect future increases in the efficacy of embryo selection.
If I just kind of blindly extrapolate based on the heritability of height in their study and what I know about the heritability of height vs intelligence in other studies, I conclude that the RDR method would generate an estimate of the heritability of intelligence of about 35-40%, which sounds pretty low, but not absurd.
It seems pretty likely we could still get a predictor that explained 30% of variance in the trait just using SNP arrays and large sample sizes, so my estimate of the efficacy of embryo selection for intelligence in the future doesn’t change much based on that.
The supplement is mostly a verbal argument along the lines of “there are just huge number of rare variants, each of which exists in only a few people, so it will be difficult to figure out how to use them to improve phenotype prediction.”
This is true, but from everything I’ve read this seems to not matter that much for one simple reason: rare alleles with small effects don’t account for that much of the variance in trait values. Like if the RDR method’s estimate of heritability is accurate for height, we should expect rare variants to raise the variance explained from 45% to 55%, which would result in a ~10% increase in expected height if you were selecting embryos based on that trait.
That would be nice, but it’s not necessary for embryo selection to work.
If your friend wants to talk about this more I’d be happy to have a conversation with them.
I have a friend who’s a geneticist and is much more pessimistic about the expected impact of embryo selection on a given desired trait. I chatted with him and read some papers he suggested and I now have a lower estimate of the expected value of embryo selection than before.
Very curious to hear from other experts in this field who disagree, especially those who don’t have a financial conflict of interest!
It seems it’s quite unclear how to even measure how heritable a given trait is, with different methods giving wildly different answers. And even if a trait is highly heritable:
The variation between siblings (or rather embryos from the same parents) is likely to be quite a lot lower than the variation seen in the general population.
The variation that’s due to very rare genetic variants is likely not captured in today’s polygenic risk scores.
Some relevant papers, which I don’t claim to fully understand, but from which I’ve extracted the above as “the gist”:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6130754/
Key thing there is probably table 2, which shows different estimates of heritability from different methods.
https://pubmed.ncbi.nlm.nih.gov/32997097/
Key quote:
“Therefore, the so-called SNV-based heritability gives the upper limit of the variance between people in their liability to a disease that can be explained by PRS and represents the variance explained by common DNA variants. As GWAS sample sizes increase, the variance explained by PRS will also increase and approach the SNV-based heritability. The SNV-based heritability estimates vary across diseases, but an approximate upper limit is approximately 30%. Although in principle, use of whole-genome sequence data could increase the variance explained by PRS (because more variance would be tagged by measured markers, ie, the SNV-based heritability approaches the heritability), it is unlikely (at least in the short term) to improve PRS (eAppendix in the Supplement). Risk stratification based on current and future PRS is illustrated in Figure 4.”
The supplement is mostly a verbal argument along the lines of “there are just huge number of rare variants, each of which exists in only a few people, so it will be difficult to figure out how to use them to improve phenotype prediction.”
Yes, I hear criticisms like those made by your friend all the time. They aren’t particularly discouraging to me.
The variation is not that much lower. The standard deviation of any continuous trait is about 70% that of the general population. That’s still plenty for selection.
The impact from rare genetic variants is mostly NOT captured in today’s genetic predictors, and that’s a place where the field could improve more in the future. But you don’t NEED rare genetic variants for embryo selection to work. You just need a predictor that correlates strongly enough with the actual trait for selection. And for many traits we already have that.
The paper you linked showing heritability is lower using relatedness disequilibrium regression was interesting. They don’t include intelligence or most of the disease predictors on which embryos are currently selected in the table, so it’s hard to give a good estimate of how strongly we should expect it to affect future increases in the efficacy of embryo selection.
If I just kind of blindly extrapolate based on the heritability of height in their study and what I know about the heritability of height vs intelligence in other studies, I conclude that the RDR method would generate an estimate of the heritability of intelligence of about 35-40%, which sounds pretty low, but not absurd.
It seems pretty likely we could still get a predictor that explained 30% of variance in the trait just using SNP arrays and large sample sizes, so my estimate of the efficacy of embryo selection for intelligence in the future doesn’t change much based on that.
This is true, but from everything I’ve read this seems to not matter that much for one simple reason: rare alleles with small effects don’t account for that much of the variance in trait values. Like if the RDR method’s estimate of heritability is accurate for height, we should expect rare variants to raise the variance explained from 45% to 55%, which would result in a ~10% increase in expected height if you were selecting embryos based on that trait.
That would be nice, but it’s not necessary for embryo selection to work.
If your friend wants to talk about this more I’d be happy to have a conversation with them.