Exploring Similarities and Differences Between Methods That Exploit Patterns of Local Genetic Correlation to Identify Shared Causal Loci Through Application to Genome‐Wide Association Studies of Multiple Long Term Conditions

TL;DR

This paper compares different methods for identifying shared genetic regions between diseases, finding that some methods are more reliable than others.

Contribution

The paper evaluates and compares the performance of several statistical tools for identifying local genetic correlations between diseases.

Findings

LAVA and ρ-HESS were found to be the most reliable in identifying genuine shared genetic factors.

A new tool, HDL-L, also showed strong performance in detecting shared genetic regions.

Results from the methods were inconsistent, with some regions identified by multiple tools and others by only one.

Abstract

Genetic correlation analysis can provide useful insight into the shared genetic basis between traits or conditions of interest. However, most genome‐wide analyses only inform about the degree of global (overall) genetic similarity and do not identify the specific genomic regions that give rise to this similarity. Identification of the key genomic regions contributing to shared genetic correlation between traits could allow the genes in these regions to be prioritised for investigation of potential shared biological mechanisms. In recent years, several statistical tools (e.g. LAVA, ρ‐HESS, SUPERGNOVA and LOGODetect) have been developed to investigate local (in contrast to global) genetic correlation. These tools partition the genome into multiple segments and provide estimates of the genetic correlation captured by each individual segment. We applied these tools to publicly available European ancestry genome‐wide association study (GWAS) summary statistics for three pairs of commonly occurring conditions: hypertension with atrial fibrillation and flutter, hypertension with chronic kidney disease, and hypertension with type 2 diabetes. Despite each of the methods aiming to address the same question, the results were found to be inconsistent across tools, with some identified regions overlapping and others implicated only by a single tool. Computer simulations using genetic data from UK Biobank, carried out under known generating conditions, suggest that LAVA and, to a lesser extent, ρ‐HESS, provide the most reliable identification of genuine shared genetic factors. A newly‐developed tool, HDL‐L, also performed highly competitively. Here we highlight the similarities and differences between the results obtained from these methods and discuss some potential reasons underlying these differences.

Figures