Genetic prediction: More data does not guarantee better accuracy

A Google study shows that European data transfer helps small Japanese cohorts, but can become unfavorable as they grow.

On September 3, Google Research presented a comparison of genetic prediction methods using UK Biobank and Biobank Japan. The researchers look at eight clinical characteristics, including blood pressure, blood lipids, and blood glucose, by varying the sizes of training populations.

European data provide an advantage when the target population is poorly documented. This advantage can disappear, and then reverse, when sufficient Japanese data are available. The threshold depends on the character studied. The result underlines the need to validate the models in the populations for which they are intended, without confusing the volume of data with clinical relevance.