An enhanced framework for local genetic correlation analysis

Abstract Genetic correlation is a key parameter in the joint genetic model of complex traits, but it is usually estimated on a global genomic scale. Understanding local genetic correlations provides more detailed insights into the shared genetic architecture of complex traits. However, LAVA, as the state-of-the-art tool for local genetic correlation analysis, reports biased statistics and, therefore, is prone to false inference. We extend the high-definition likelihood (HDL) method to a local version, HDL-L (HDL-Local), which divides the genetic correlation analysis into semi-independent loci. HDL-L allows for a more granular estimation of genetic variances and covariances. Simulations show that HDL-L offers more consistent heritability estimates and more efficient genetic correlation estimates compared to LAVA. Across extensive simulations under different heritability settings, HDL-L maintained robust performance. In the analysis of 30 phenotypes from the UK Biobank, HDL-L identified 889 significant local genetic correlations across 658 loci, while LAVA identified 696 significant estimates across 441 loci. Furthermore, HDL-L demonstrated a significant computational advantage, being around 50 times faster than LAVA in the simulations. HDL-L proves to be a powerful tool for uncovering the detailed genetic landscape that underlies complex human traits, offering both superior accuracy and computational efficiency.