bio-causal-genomics-colocalization-analysis

What it does

The skill teaches how to test whether two or more traits share a causal variant at a locus using Bayesian colocalization methods. It covers methods including coloc.abf, coloc.susie, HyPrColoc, moloc, eCAVIAR, SMR/HEIDI, PWCoCo, and SharePro_coloc, and outlines when to apply each approach to integrate GWAS with QTL or molecular QTL data. It provides guidance on computing posterior probabilities for five hypotheses (H0, H1, H2, H3, H4) to distinguish true causal overlap from LD-driven coincidence, and emphasizes running sensitivity analyses over p12 priors and PP.H4 thresholds.

How it works

• It enumerates multiple methods and their inputs/outputs, showing concrete commands and workflows: coloc.abf using dataset1 and dataset2 with p12, coloc.susie via per-credible-sets, HyPrColoc with effect.est and effect.se matrices, SMR with HEIDI via CLI, eCAVIAR for CLPP per SNP, and PWCoCo following GCTA-COJO conditioning.
• It presents an algorithmic taxonomy table detailing model assumptions, inputs, outputs, strengths, and failure modes for each method.
• It includes a decision tree by scenario recommending methods (e.g., coloc.abf for single-causal loci, coloc.susie or PWCoCo when allelic heterogeneity is suspected, HyPrColoc for many-trait clustering, moloc or HyPrColoc for multi-omics or multi-trait setups).
• It outlines per-method failure modes (e.g., LD-reference mismatch for coloc.susie, trans-eQTL priors for coloc.abf, MHC/HLA-region caveats, lead-SNP window bias, underpowered eQTL data) and provides fixes or workarounds.
• It provides a PP.H4 threshold framework and guidance to report PP.H3 alongside PP.H4, along with a reminder to run coloc::sensitivity and report the p12 range where PP.H4 remains above thresholds.
• It cautions on reference panel choices (GTEx v8 vs v10, eQTLGen), and notes operational adjustments like lowering p12 for trans-eQTL pairs and avoiding single-causal assumptions in long-range LD regions.

When to use it

• When testing whether GWAS signals colocalize with eQTL, sQTL, pQTL, or mQTL signals across one or more traits.
• When distinguishing shared causal variants from LD-driven coincidence, especially with allelic heterogeneity or multiple independent signals.
• When choosing between single-causal vs multi-causal methods, or when harmonising summary statistics for colocalization.
• When performing sensitivity analyses over prior assumptions (p12) and thresholds for PP.H4.

What it can touch

• Tools and commands explicitly named: coloc (R package), coloc.abf, coloc.susie, HyPrColoc, SMR, HEIDI, eCAVIAR, PWCoCo, GCTA-COJO, and related CLI/R commands as shown in the examples.
• It references specific version checks and commands to verify installed versions (packageVersion('coloc'), smr --version, pwcoco --help, sharepro_coloc.py --help).
• It recognizes data inputs like beta+varbeta, p+MAF, z-scores, LD matrices, and per-trait summary stats, as required by the listed methods.

Caveats

• It notes limitations such as single-causal assumptions in coloc.abf failing with multiple signals, LD-reference mismatches, trans-eQTL priors being too liberal with default p12, and the need to treat MHC/HLA regions with specialized approaches (HLA-coloc or MHC exclusion).
• It warns about computational and interpretive challenges for high-dimensional multi-trait analyses (moloc scaling, AH handling in HyPrColoc).
• It emphasizes the need to anchor version compatibility (coloc 5.2.3+, susieR 0.12.35+, hyprcoloc 1.0+, SMR 1.3.1+, eCAVIAR 2.2+, PWCoCo 1.0+, moloc 0.1+, SharePro_coloc 7.0+).
• It advises careful window definition, avoiding lead SNP-centered biases, and ensuring adequate eQTL/MQTL sample sizes (with underpowered eQTL highlighted as a limitation).