LRCP: identifying functionally related SNPs and genes from phenome-wide genetic-effect correlations using GWAS summary statistics
Genetic correlation describes which traits share genetic causes. The transposed question, which variants act alike across the phenome, has no comparable estimator, although it points directly to shared genes, pathways and regulators. Correlations of GWAS Z scores between variants across many traits are easy to compute, but they are confounded by linkage disequilibrium (LD) and, for any pair of…
The article "Identifying Functionally Related SNPs and Genes from Phenome-Wide Genetic-Effect Correlations Using GWAS Summary Statistics" introduces a method called LRCP (likelihood and regression on cross-products of Z scores) to estimate the phenome-wide genetic-effect correlation (rho) between single nucleotide polymorphisms (SNPs).
The article discusses the limitations of naive correlations due to linkage disequilibrium (LD) and sample overlap, which can lead to a high rejection rate of null pairs of distant variants. LRCP addresses these issues by regressing Z-score cross-products on LD cross components, taking into account per-SNP heritability enrichment from a companion method called LRCQ and LDSC-derived plug-ins.
The method is unbiased, with 95% confidence intervals covering 93-98% and a type-I error rate of 5.0-5.7%. In simulations, LRCP accurately identifies true correlations while keeping false positives low. In the Pan-UK Biobank dataset, LRCP recovers the expected sign relationships among PCSK9, LDLR, and APOB variants, and estimates the correlation between the TP53 polyadenylation variant rs78378222 and MDM2 rs3730556 as 0.90.
This correlation occurs within a shared body-size and red-cell trait axis. The article highlights the precision limitations of current biobanks, which are determined by the number of genetically distinct traits. The researchers demonstrate that among 7,793 enriched LD clumps containing 30.2 million distal pairs, LRCP identifies 17 pairs with a false discovery rate (FDR) of 5%, while the naive correlation method flags 23 million pairs.
The article concludes that LRCP is a conservative and effective tool for identifying functional relationships between SNPs in a phenome-wide context.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
