Genetic risk prediction across diverse populations

The vast majority of GWAS are performed in individuals of European descent. The applicability of these genetic findings to non-European populations varies with genetic divergence, differences in LD and allele frequencies, and... [ view full abstract ]

The vast majority of GWAS are performed in individuals of European descent. The applicability of these genetic findings to non-European populations varies with genetic divergence, differences in LD and allele frequencies, and genetic architecture. Human history provide a critical lens into complex trait studies, including the generalizability of genetic risk prediction to understudied populations. By simulating genetic data that models human history, we have shown that genetic risk predicted using European summary statistics transfers poorly to non-European populations. We have also empirically evaluated genetic risk prediction across populations using results from the Psychiatric Genetics Consortium. We find that East Asian schizophrenia risk is better predicted by summary statistics from East Asian cohorts (13k cases and 16k controls) than from ~3-fold larger European cohorts (37k cases and 113k controls, Nagelkerke’s R² = .104 vs .066). To improve cross-population genetic risk prediction, we are developing novel statistical methods to improve prediction accuracy across populations, such as when GWAS summary statistics are available from multiple populations. This method models LD structure in each respective population to better approximate causal effect sizes used in prediction. Our work cautions that findings from large-scale GWAS may have limited generalizability across populations with standard approaches, highlighting the need to include more diverse individuals in medical genomics.