Accurate Liability Estimation Substantially Improves Power in Ascertained Case Control Studies

Omer Weissbrod, Christoph Lippert, Dan Geiger, David Heckerman
(Submitted on 8 Sep 2014)

Future genome wide association studies (GWAS) of diseases will include hundreds of thousands of individuals in order to detect risk variants with small effect sizes. Such samples are susceptible to confounding, which can lead to spurious results. Recently, linear mixed models (LMMs) have emerged as the method of choice for GWAS, due to their robustness to confounding. However, the performance of LMMs in case-control studies deteriorates with increasing sample size, resulting in reduced power. This loss of power can be remedied by transforming observed case-control status to liability space, wherein each individual is assigned a score corresponding to severity of phenotype. We propose a novel method for estimating liabilities, and demonstrate that testing for associations with estimated liabilities by way of an LMM leads to a substantial power increase. The proposed framework enables testing for association in ascertained case-control studies, without suffering from reduced power, while remaining resilient to confounding. Extensive experiments on synthetic and real data demonstrate that the proposed framework can lead to an average increase of over 20 percent for test statistics of causal variants, thus dramatically improving GWAS power.