The Same Polygenic Scores That Prove Rapid Selection Warn Against Reading Group Differences
2026-07-21 · AI-generated insight
Put these two notes side by side and they lean on the same instrument while pulling it in opposite directions.
The ancient DNA paper offers Warne something he wants: proof that trait-relevant allele frequencies can shift fast. Over just ten millennia, hundreds of alleles underwent strong directional selection in West Eurasians, producing standard-deviation-scale changes in the polygenic predictors of complex traits—including, the authors note, increases in "measures of cognitive performance." That establishes the biological plausibility Warne needs: selection acting on standing variation can, in principle, move populations apart on polygenic traits within evolutionary eyeblinks. It undercuts any claim that group-level genetic differences in a polygenic trait are impossible on timescales this short.
But the same paper carries a caveat that lands squarely on Warne's fourth line of evidence—polygenic scores applied to diverse samples. Reich and colleagues are careful: these effects "were measured in industrialized societies, and it remains unclear how these relate to phenotypes that were adaptive in the past." The score is not the trait; it is a correlational shadow calibrated in one environment, whose meaning outside that environment is genuinely uncertain. That is precisely the portability problem that makes cross-group polygenic comparison treacherous. GWAS effect sizes are estimated within a population and confounded with its structure; carrying them across populations or across environments is the contested move, not a settled one.
So the geneticists whose method most vividly demonstrates that populations can diverge under selection are also the ones flagging why the polygenic-score readout can't be taken at face value across contexts. Warne's admixture and Spearman's-hypothesis arguments don't depend on this, but his genomic line does—and the ancient DNA authors, working the same data with far larger samples, treat the industrialized-society calibration as a live limitation rather than a footnote.
The honest synthesis: mechanism is plausible, measurement is not yet trustworthy. The tool that best shows selection is real is the same tool that isn't ready to tell us where any given group sits.