Genome-wide iterative fine-mapping for non-Gaussian phenotypes.

Fine-mapping seeks to identify causal variants associated to a phenotype of interest. Fine-mapping searches through genomic regions previously identified by single marker analysis of genome-wide association studies (GWAS) data. This two-stage approach (1) often fails to identify causal variants with smaller effect sizes, and (2) does not properly correct for multiple comparisons. The former problem leads fine-mapping to have low recall of causal genetic variants, and the latter problem leads to high false discovery rate (FDR). To address these issues, we propose a novel Genome-wide Iterative fiNe-mApping method for non-Gaussian data (GINA-X). GINA-X efficiently extracts information from GWAS data by iterating a screening step and a variable selection step. The screening step provides a list of candidate genetic variants and an estimate of the proportion of null genetic variants. After that, the variable selection step selects a more focused list of genetic variants using the estimated null proportion to appropriately control for genome-wide multiplicity. A simulation study shows that, when compared to competing fine-mapping methods, GINA-X reduces FDR and increases recall. Case studies on alcohol use disorder and breast cancer show that GINA-X provides more focused lists of candidate causal genetic variants with better predictive performance.