Improving fine-mapping by modeling infinitesimal effects

Abstract Fine-mapping aims to identify causal variants for phenotypes. Bayesian fine-mapping algorithms (e.g.: SuSiE, FINEMAP, ABF, and COJO-ABF) are widely used, but assessing posterior probability calibration remains challenging in real data, where model misspecification likely exists, and true causal variants are unknown. We introduce Replication Failure Rate (RFR), a metric to assess fine-mapping consistency by down-sampling. SuSiE, FINEMAP and COJO-ABF show high RFR, indicating potential under-conservative mis-calibration. Simulations reveal that non-sparse genetic architecture can lead to miscalibration, while imputation noise, non-uniform distribution of causal variants, and QC filters have minimal impact. We present SuSiE-inf and FINEMAP-inf, novel fine-mapping methods modeling infinitesimal effects alongside fewer larger causal effects. Our methods exhibit improved calibration, RFR and functional enrichment, competitive recall and computational efficiency. Notably, using our methods’ posterior effect sizes substantially increases PRS accuracy over SuSiE and FINEMAP. Our work improves causal variants identification for complex traits, a fundamental goal of human genetics.