Transfer learning for genomic prediction in underrepresented populations

Transfer learning from European cohorts improves prediction in small underrepresented populations but degrades accuracy in larger cohorts.

Researchers from Google Research evaluated methods to enhance cross-population genetic risk prediction.

They found that transfer learning from European cohorts can improve prediction in small underrepresented populations.

However, this improvement is short-lived and accuracy degrades once the target cohort sample sizes grow, particularly for traits with population-specific genetic architectures.