Suppose that the genetic architecture of MD consists of many small-effect loci, smaller than can be detected at genome-wide significance by currently available samples. For example, suppose the odds ratio for these risk variants are 1.05 and suppose the variants have a frequency of 50% (alleles with a higher frequency are easier to detect, so this is a conservative assumption). Power to detect a single variant of this effect size at this frequency in a sample size of 10,000 cases and 10,000 controls is less than 0.001%, at a p value of 1 × 10−7 and disease prevalence of 10% (Purcell et al., 2003). But there is a 67% chance that such a variant will have a p value less than 0.5. This means that if all SNPs are ranked by their p values, then p values less than 0.5 will be enriched with SNPs that contribute to disease susceptibility. In other words, if there are small-effect variants contributing to MD, then the distribution of SNP p values will depart from null expectations. This method is referred to as polygenic scoring and has been used to investigate the polygenic nature of complex traits.
