Cost-effective non-additive GWAS across 2329 diseases in 500,349 individuals

Drug candidates supported by genetic evidence are more likely to succeed in clinical trials, with genome-wide association studies (GWAS) providing a key source of such evidence. Standard GWAS approaches assume additive effects of alleles on the phenotype, but non-additive models have also successfully identified novel associations across various traits. Despite their potential, the large-scale application of non-additive GWAS across thousands of phenotypes in biobanks has been limited by high computational costs. To address this challenge, we present a method that leverages the correlation between additive and non-additive p-values to prioritize variants likely to reach genome-wide significance in non-additive analyses. Applied to the FinnGen dataset comprising 500,349 individuals and 2329 phenotypes, this method reduces computational costs by three orders of magnitude while retaining nearly all true non-additive associations, identifying 781 novel loci missed by additive GWAS. We report fine-mapping and colocalization with 571 datasets for novel loci, uncovering likely causal variants and potential insights into biological mechanisms.