An alternate approach to assessing the genetic overlap between two traits is to estimate the correlation between causal effect sizes across the two traits. Genome-wide genetic correlations can be estimated from individual-level data using bivariate REML101. A recent study estimated genome-wide genetic correlations from summary data using the information in polygenic risk scores, although this approach required LD-pruning the data which may lead to upwards bias84. Another recent study estimated genome-wide genetic correlations from summary data using cross-trait LD score regression102, which generalizes LD score regression to regress products of z-scores against LD scores for each SNP; this method produced estimates that were highly concordant with those from individual-level data101. Fitting the underlying MVN model using maximum likelihood instead of linear regression has produced promising results in applications to estimating cross-trait and cross-population genetic correlations, and may also prove useful in other settings103. Although genetic correlation analyses restricted to associated variants have also produced important findings97, the power of methods that leverage polygenic signals in genome-wide data is underscored by the discovery of significant genetic correlations involving traits with zero
