A heritability-optimized method for functional prioritization of rare coding variants in complex traits

Abstract

Population sequencing studies are uncovering millions of missense rare variants (RVs) implicated in the ‘missing heritability’ of complex diseases, yet distinguishing functional from neutral variation remains challenging. Here we present rare variant heritability-optimized scores (RovHer), a heritability-optimized method for inferring a variant’s probability of being functional (that is, its contribution to trait variance). RovHer leverages exome-wide association statistics across 4,927,334 RVs and uses multivariate adaptive regression splines to integrate 75 variant- and gene-level features without a priori constraints. Across 21 traits in up to 357,086 UK Biobank European individuals, the top 1% of RovHer-prioritized missense RVs (m = 13,410) explained, on average, 16.1% of total RV heritability, a 10-fold gain over 7 alternative methods (average 1.5%; range 1.2–2.3%). RovHer’s predictions are independent of allele frequency, generalize in a multi-ancestry replication and effectively identify high-risk carriers of severe clinical outcomes across 17 monogenic gene–trait pairs. Our tool showcases the value of missense RVs in population risk stratification. This study presents RovHer, a heritability-optimized scoring method for estimating the functional impact of rare variants, which is applied to the UK Biobank to demonstrate its ability to enrich for missense rare variants that maximally explain complex trait heritability.