Logica: A likelihood framework for cross-ancestry local genetic correlation estimation using summary statistics

Understanding genetic architecture across ancestries through genetic correlation analysis is critical for determining the degree to which genetic factors underlying diseases or complex traits are shared or differ among populations. Current methods for genetic correlation analysis primarily rely on method of moments approaches and focus on estimating the global genetic correlation across the entire genome. However, these methods often overlook important local genomic complexities and inadequately model the intricate linkage disequilibrium (LD) structures that vary substantially across ancestries. Here, we present Logica (local genetic correlation across ancestries), a method specifically designed to estimate local genetic correlations across ancestries and in admixed populations. Logica employs a bivariate linear mixed model that explicitly accounts for diverse LD patterns across ancestries, operates on genome-wide association study summary statistics, and utilizes a maximum-likelihood framework for robust inference. An important by-product of Logica is a joint heritability test across ancestries that yields well-calibrated p values-an aspect that existing approaches often struggle with. We conducted comprehensive evaluations of Logica through realistic simulations and analyses of 13 complex traits from multiple biobanks. Simulations showed that Logica achieves improved accuracy in local genetic correlation estimation (with mean squared errors 2.23-4.13 times lower) and enhanced power for detecting genetically correlated regions (8%-40% increase with controlled false discovery rate [FDR] at 5%). In real data, Logica produced valid genetic correlation estimates across all genomic regions, whereas existing methods failed in 23%-39% of regions. Additionally, Logica exhibited better FDR control (14%-58% improvement), identifying genetically correlated regions with greater functional relevance.