Given a vector of estimated effect sizes b^i from a genome-wide association study performed on a set of training samples, the polygenic risk score (International Schizophrenia Consortium et al., 2009) (PRS) for a target individual with genotypes gi is defined as y^=∑i=1Mbl^gi. In practice, rather than computing the PRS using estimated effect sizes for all available genetic markers, the PRS is computed on a subset of genetic markers obtained via informed LD-pruning (Stahl et al., 2012) (also known as LD-clumping) followed by P-value thres holding (International Schizophrenia Consortium et al., 2009). Specifically, this “pruning + thres holding” strategy has two parameters, RLD2 and PT, and proceeds as follows. First, we prune the SNPs based on a pair wise threshold RLD2, removing the less significant SNP in each pair (using PLINK; see Web Resources). Second, we restrict to SNPs with an association P-value below the significance threshold PT.
