Improving fine-mapping by modeling infinitesimal effects
02:13 Share

Link to bioRxiv paper: http://biorxiv.org/cgi/content/short/2022.10.21.513123v1?rss=1

Authors: Cui, R., Elzur, R. A., Kanai, M., Ulirsch, J. C., Weissbrod, O., Daly, M., Neale, B., Fan, Z., Finucane, H. K.

Abstract: Fine-mapping aims to identify genetic variants that causally impact a given phenotype. State-of-the-art Bayesian fine-mapping algorithms (for example: SuSiE1, FINEMAP2,3, ABF4, and COJO5-ABF) are widely applied in practice6-11, but it remains challenging to assess their calibration (i.e., whether or not the posterior probability of causality reflects the true proportion of causal variants) in real data, where model misspecification almost certainly exists and true causal variants are unknown. Here, we present the Replication Failure Rate (RFR), a metric to assess the consistency of fine-mapping results based on downsampling a large cohort. Empirical evaluation of fine-mapping results from SuSiE, FINEMAP and COJO-ABF suggest that these methods may be miscalibrated in the under-conservative direction. Next, we show in simulations that non-sparse genetic architecture can lead to miscalibration, while imputation noise, non-normal effect size distributions, and quality control filters removing potentially causal variants are less likely contributors. Here, we present two new fine-mapping methods, SuSiE-inf and FINEMAP-inf, that extend SuSiE and FINEMAP to incorporate a term for infinitesimal effects in addition to a small number of larger causal effects of interest. Our methods exhibit better calibration in simulations and improved RFR and functional enrichment in real data, with minimal loss of recall and competitive computational cost. Furthermore, using the sparse fine-mapped variants identified by our methods to perform cross-population genetic risk prediction in the UK Biobank, we observed a substantial increase in predictive accuracy over SuSiE and FINEMAP. Our work improves our ability to pinpoint causal variants for complex traits, a fundamental goal of human genetics.

Copy rights belong to original authors. Visit the link for more info

Podcast created by Paper Player, LLC