An inherent limitation of genetic correlation estimates using LD score regression is that they capture only shared common genetic variation between two phenotypes, and therefore the estimates are typically much lower than traditional measures of heritability. For example, the estimated heritability from twin studies of nicotine dependence is 59% (Li, 2006), relative to our GWAS heritability estimate of 15%, and the estimated heritability of smoking initiation is 50% (Li et al., 2003), relative to our GWAS heritability estimate of 8%. The discrepancy between traditional heritability measures and GWAS heritability is due to the fact that these estimates are based on GWAS data, filtered to include only SNPs with minor allele frequency greater than 0.01. In addition, the effects of the SNPs must all be in the same direction; SNPs that contribute to both phenotypes, but have opposite effect sizes will reduce the genetic correlation estimates. These limitations, however, do not bias the methodology, but instead reduce the power of the methodology. Therefore, the observed associations are likely sound.
