QCAT: testing causality of variants using only summary association statistics

Abstract

Genome-wide and, very soon, sequencing association studies, might yield multiple regions harbouring interesting association signals. Given that each region encompasses numerous variants in high linkage disequilibrium, it is not clear which are i) truly causal or ii) just reasonably close to the causal ones. Researchers proposed many methods to predict, albeit not test, the causal SNPs in a region, a process commonly denoted as fine-mapping. Unfortunately, all existing fine-mapping methods output posterior causality probabilities assuming that causal SNPs are among those already measured in the study, or have been catalogued elsewhere. However, due to technological and computational obstacles in calling many types of genetic variants, such assumption is not realistic. We propose a novel method/software, denoted as Quasi-CAausality Test (QCAT), for testing (not just predicting) the causality of any catalogued genetic variant. QCAT i) makes no assumption that causal variants are among catalogued variants, and ii) makes use of easily available summary statistics from genetic studies, e.g. variant association Z-scores, to make statistical inferences. The proposed statistical test controls the type I error at or below the desired level. Its practical application to well-known smoking association signals provide some insightful results. The QCAT software is publically available at: http://dleelab.github.io/qcat/