Emil O W Kirkegaard

"Given currently available polygenic predictors" "𝑟2_ps=4.3%" ^ There's the problem. *Current* polygenic scores are shit. Someone more knowledgeable like @KirkegaardEmil feel free to correct me. What the prospects for improving r2_ps is beyond my understanding...

... but I understand that they improved a lot fairly recently & the sample size of 230,000 for IQ is still 3 orders of magnitude smaller than what we can expect with population level data collection. @gwern has written on this: https://www.gwern.net/Embryo-selection#gwas-improvements …

so reading @gwern I get the impression that the limiting factor here is poor quality data about traits & genes, utimately because we're not really trying very hard to do it properly. Maybe with falling prices for whole genome seq and privacy preserving tech we can improve?

I'd say it's more limited by analysis (eg why do they consider only single trait selection when they know perfectly well that understates the value by a factor of >4?) and number of embryos (which they also ridiculously downplay).

There is a difference between credible (people are willing to believe) and plausible (has actual support) scenarios. Credible scenarios are a subset of plausible scenarios when we factor out credible but wrong scenarios. However, credible scenarios are more publishable.