Polygenic Scores: Why The Hype?
 |
| A single number, thousands of variants — but does the promise match the science? (📷:empowervmedia) |
Imagine a single number that claims to predict your risk of depression, your expected years of schooling, or your likelihood of a heart attack — derived entirely from your DNA. That is the promise of the polygenic score (PGS): a summary statistic aggregating the influence of thousands of common genetic variants into one figure. Since their first systematic development in 2009, over 1,000 peer-reviewed publications have employed this methodology - spanning schizophrenia, educational attainment, and alcohol misuse. The scientific appeal is genuine. The hype surrounding it, however, demands scrutiny.
Polygenic scores are statistical predictors, not biological verdicts. Within the populations for which they were built, their explanatory power can be notable - reaching up to 13% of variance explained for educational attainment and 24% for height (Mostafavi et al., eLife, 2020). Yet that accuracy collapses the moment a score crosses an ancestry boundary. Because scores built on predominantly European reference datasets perform poorly in individuals of non-European ancestry (Martin et al., PMC, 2024), a deep structural inequity runs through the entire enterprise: as of early 2024, 95% of genome-wide association study participants were of European descent - with only 0.2% of African ancestry (GWAS Diversity Monitor, cited in Dareng et al., 2025). In medicine, that asymmetry is not a footnote. It is a patient-safety issue.
"Polygenic scores are statistical predictors, not biological verdicts. Their predictive accuracy can drop sharply (from ~13% to as little as 0.2% of variance explained) when applied across ancestry groups."
The clinical promise is nonetheless real. In cardiovascular medicine, multi-ancestry polygenic risk score frameworks have demonstrated improved risk stratification beyond traditional clinical factors, using datasets of over 225,000 participants (Nutrients, 2025). Similarly, recent reviews highlight genuine progress in disease prediction and preventive strategy - alongside unresolved ethical challenges around ancestry bias, privacy, informed consent, and clinical interpretation (iScience, Elsevier, 2025). The danger lies not in the methodology per se, but in its overreach: the casual collapse of statistical correlation into causal explanation, and the downstream misapplication of scores in educational sorting, insurance underwriting, or direct-to-consumer genomics.
What we are witnessing is a familiar pattern in the history of science communication: a genuinely promising methodology amplified into a near-universal solution by media coverage, venture capital, and the seductive simplicity of a single number. The nuanced peer-reviewed literature - on gene–environment interactions, social determinants of health, and the hard limits of heritability (Personalized Environment and Genes Study, PMC, 2025) - rarely makes the headline. The full article unpacks the science, interrogates the ethics, and poses the question the hype machine would rather you didn't ask: a statistical association with what, for whom, and at whose expense?
| → Click here for full article |
⭐⭐⭐
