Here we illustrate the use of PRSice to test for shared genetic aetiology between traits. We first investigate the genetic relationship between schizophrenia (SCZ) and major depressive disorder (MDD), both known to be complex and comorbid. We apply PRSice to replicate the finding by Smoller et al. (2013) that SCZ PRS can predict MDD status, using the RADIANT-UK MDD case-control data set (Supplementary Note S2, Lewis et al., 2010). Applying PRSice, we remove SNPs in linkage disequilibrium and include principal components to control for population structure. We find significant evidence that SCZ PRS predict MDD status, and under the approach of only testing PRS at several broad P-value thresholds find the most predictive threshold at PT = 0.05 (Fig. 1). Next we repeat the analysis using high-resolution PRS (Supplementary Note S3) and find the most predictive PRS at PT=0.0265 (Fig. 2). The PRS at PT=0.05 explains 1.5% of the variation in MDD (Nagelkerke R2; P=1.3×10−9) whereas the high-resolution best-fit PRS explains 2.1% (P=2.1×10−12) and is based on 5252 fewer SNPs (12148 rather than 17400).
