Going through multiple generations without development of a full organism, selecting only on what are thought to be good traits in the genotype, seems like it would be vulnerable to mutations (or just combinations of existing alleles) that make the full organism non-viable, or weak.
For "lower" organisms, I guess one could try it and see how much of a problem this is, but for humans it seems like this is an ethical problem with this technique (hardly the only one, of course).
This seems like a good way to max out our IQ given our current knowledge of genetic predictors of IQ. But how much gain is there without having to discover more genetic predictors? Three standard deviations?
After that we would need new ones and discovery of new genetic predictors seems slow as you have to wait for maturation of an embryo to a human (as embryo's can't take IQ tests).
Ok, so maybe I'm being really dumb here, but I see an issue with this approach:
If you're optimizing for <x> based on polygenic scores, you can't really optimize past the known limits of that trait. As in, you might have a given set of traits that optimize for <x> which are not naturally present in any genome of that species all together, but the chances of all of them being present actually optimizing for <x> is minor.
For example:
You can knockout certain types of IGF1 or IGF2 and that would increase longevity. You can partially disable IGF receptors, again, longer lifespan. But knockout IGF1 and 2, and select all the disabling mutations on IGFRs and you get... probably an organism that won't last for long, if it even gets past the tissue differentiation stage.
You can partially disable myostatin activity via mutations in one or both copies and optimize muscle growth. You can select for various traits that select for testosterone production and optimize muscle growth. You can select for various traits that increase androgen receptor density in certain tissues and optimize muscle growth. But do all of them combined and... you probably get an organism that will soon die after it's bones get crushed under the strength and weight of it's own muscles, if not worst.
More broadly, once you reach a genome that's divergent enough from any known genomes, polygenic scores are of no use, since they operate well inside the realm of low-variance (comparative to this method) 21st century human genomes.
So it seems unlikely that optimizing for <x> via polygenic scores will actually have the desired result past a certain threshold of "known" configurations... At which point, why not simply get iPSCs using tissue from the individual with said configuration?
Further more, even if the risk of this technique exists but it has better-than-random chance at optimizing for <x>, it wouldn't be viable for species where there are ethical issues and/or for phenotypes that take long to manifest. Since validating it is hard.
If ethical and development-cost issues were non-existent or minimal, why not select for the trait via some sort of environment that encourages optimizing that trait, which would likely result in exploring a much wider mutation-spaces and unusual combinations which polygenic scoring wouldn't reveal as beneficial for <x>. As is already done with e.g. viruses and bacteria.
I'm writing this at 5AM and have only surface-level understanding of the subject, so I'm 99% certain I'm missing some nuance, but I'd love to understand what said nuance is, I've been scratching my head unable to find it.
Going through multiple generations without development of a full organism, selecting only on what are thought to be good traits in the genotype, seems like it would be vulnerable to mutations (or just combinations of existing alleles) that make the full organism non-viable, or weak.
For "lower" organisms, I guess one could try it and see how much of a problem this is, but for humans it seems like this is an ethical problem with this technique (hardly the only one, of course).
Any chance you want to do a slightly more in-depth write-up to guest author on my substack?
This seems like a good way to max out our IQ given our current knowledge of genetic predictors of IQ. But how much gain is there without having to discover more genetic predictors? Three standard deviations?
After that we would need new ones and discovery of new genetic predictors seems slow as you have to wait for maturation of an embryo to a human (as embryo's can't take IQ tests).
Ok, so maybe I'm being really dumb here, but I see an issue with this approach:
If you're optimizing for <x> based on polygenic scores, you can't really optimize past the known limits of that trait. As in, you might have a given set of traits that optimize for <x> which are not naturally present in any genome of that species all together, but the chances of all of them being present actually optimizing for <x> is minor.
For example:
You can knockout certain types of IGF1 or IGF2 and that would increase longevity. You can partially disable IGF receptors, again, longer lifespan. But knockout IGF1 and 2, and select all the disabling mutations on IGFRs and you get... probably an organism that won't last for long, if it even gets past the tissue differentiation stage.
You can partially disable myostatin activity via mutations in one or both copies and optimize muscle growth. You can select for various traits that select for testosterone production and optimize muscle growth. You can select for various traits that increase androgen receptor density in certain tissues and optimize muscle growth. But do all of them combined and... you probably get an organism that will soon die after it's bones get crushed under the strength and weight of it's own muscles, if not worst.
More broadly, once you reach a genome that's divergent enough from any known genomes, polygenic scores are of no use, since they operate well inside the realm of low-variance (comparative to this method) 21st century human genomes.
So it seems unlikely that optimizing for <x> via polygenic scores will actually have the desired result past a certain threshold of "known" configurations... At which point, why not simply get iPSCs using tissue from the individual with said configuration?
Further more, even if the risk of this technique exists but it has better-than-random chance at optimizing for <x>, it wouldn't be viable for species where there are ethical issues and/or for phenotypes that take long to manifest. Since validating it is hard.
If ethical and development-cost issues were non-existent or minimal, why not select for the trait via some sort of environment that encourages optimizing that trait, which would likely result in exploring a much wider mutation-spaces and unusual combinations which polygenic scoring wouldn't reveal as beneficial for <x>. As is already done with e.g. viruses and bacteria.
I'm writing this at 5AM and have only surface-level understanding of the subject, so I'm 99% certain I'm missing some nuance, but I'd love to understand what said nuance is, I've been scratching my head unable to find it.
Fascinating! I'm excited to see what comes out of this research.
Passing thought: would iterated meiotic selection or IES be improved by using gene drives to ensure that certain genes are retained in each iteration?
https://en.wikipedia.org/wiki/Gene_drive