Accounting for uncertainty in residual variances improves calibration of the Sum of Single Effects model for small sample sizes

Abstract

The Sum of Single Effects (SuSiE) model is a widely adopted method for genetic fine-mapping. We show that, in small-sample studies, the original SuSiE fitting procedure produces substantially higher rates of false positive findings. We show that a simple modification to SuSiE improves performance, especially in small-sample studies. This modification is particularly important for emerging molecular QTL applications in rare cell types and primary tissues where sample sizes are inherently limited.