Can a score built from thousands of DNA variants predict your future health?

A polygenic score is a number that summarizes the estimated effect of many genetic variants on a person's trait or disease risk. It is typically a weighted sum of trait-associated alleles, and for a risk predictor it can depend on thousands of individual variants, each with a small effect. The score reflects genetics only, without taking environmental factors into account.

How well do they work? Researchers measure performance with a number called AUC, where a larger number implies better prediction. In 2018 a coronary disease score reached about 0.64 and a 2020 schizophrenia score about 0.71. Most polygenic scores are not predictive enough to diagnose disease, although they can improve estimates of susceptibility when used alongside factors like age and smoking status, and they may be very accurate for people at the extremes: those with the top 1% of scores for cardiovascular disease had a lifetime risk above 10%, comparable to people with rare genetic variants.

There are caveats. As of 2021 most data came from populations of European ancestry, which is why the scores generally perform better in that ancestry. And genetic risk has to be read in the context of environment: a person with a high genetic risk for alcoholism is at lower risk if never exposed to alcohol.

The most contested use is embryo selection. Tests for polygenic diseases were first used on embryos in 2019, and the practice has been criticised for ethical and safety issues and limited practical utility, while others have made arguments in its favour. The debate is still open.