In order to address these limitations, Bulik-Sullivan et al previously proposed a different method, LD score regression (Bulik-Sullivan et al., 2015a). Essentially the method involves regressing summary results statistics from millions of genetic variants across the genome on a measure of each variant’s ability to tag other variants locally (i.e. its ‘LD score’). The intuition behind the approach is that if a trait is genetically influenced, then variants that tag more of the genome (i.e. have high LD scores) should have a greater opportunity to tag causal variants and therefore have higher test statistics on average than variants that have low LD scores. In this way genome-wide inflation of test statistics due to genuine polygenicity can be distinguished from biases such as population stratification and cryptic relatedness. The basic method is very flexible and can be adapted to estimate SNP heritability, calculate a more accurate and efficient genome-wide inflation correction factor than genomic control (Bulik-Sullivan et al., 2015a), partition the SNP heritability by functional category (Finucane et al., 2015), and estimate the genetic correlation between different complex traits and diseases (Bulik-Sullivan et al., 2015b), all using GWAS summary-level results data (Table 1).
