12 · Causal TWAS (cTWAS)

Overview

This pipeline runs cTWAS (causal Transcriptome-Wide Association Study) to jointly model gene expression and GWAS variants, distinguishing genes with direct evidence of causal effects on disease from those with only indirect (LD-mediated) associations. It uses the FUSION weights generated in pipeline 11 and GWAS summary statistics from pipeline 07.

Workflow Logic

graph TD
    classDef snazzy fill:#f1f1f1,stroke:#333,stroke-width:2px,color:#000;
    classDef out fill:#e1f5fe,stroke:#01579b,color:#000;

    A[1000G hg38 PLINK files] --> B(create_ld_matrices)
    C[FUSION .pos + .wgt.RDat files] --> D(copy_fusion_weights)
    D --> E(run_ctwas)
    B --> E
    F[GWAS hg38 TSV] --> E
    G[SMR 1000G BIM] --> E
    E --> H(ctwas_report)

    class A,B,C,D,E,F,G snazzy
    class H out

  1. create_ld_matrices: Pre-computes LD matrices and variant information files from the 1000G hg38 PLINK reference panel. These are shared across all cell types and GWAS, so this rule runs once.
  2. copy_fusion_weights: Copies only genuine (non-zero-byte) FUSION weight files from pipeline 11 into a clean per-cell-type directory. Zero-byte stub files — created for genes where FUSION found insufficient heritability — cause ctwas::preprocess_weights() to crash, so they must be excluded before cTWAS runs.
  3. run_ctwas: Runs the full cTWAS analysis for each cell type × GWAS combination using ctwas_run.R. Fits the cTWAS mixture model, estimates prior probabilities for SNP and gene effects, and outputs posterior inclusion probabilities (PIPs) for all genes.
  4. ctwas_report: Renders an RMarkdown HTML report summarising cTWAS gene discoveries across cell types and GWAS traits, including locus plots where available.
Note

Stub file filtering

FUSION produces zero-byte .wgt.RDat files for genes it skips (insufficient cis-SNP heritability). The copy_fusion_weights rule filters these out by reading the .pos file — which only lists genes with real weight files — and copying only those files:

awk 'NR>1 {print $2}' {input} | \
while read wgt; do
    cp -v {params.src_dir}/"$wgt" {params.dest_dir}/
done

This ensures cTWAS only sees valid weight files and avoids preprocess_weights() errors.

Key Config Parameters

# cTWAS uses the SMR 1000G reference BIM for variant information
# LD matrices are computed from ldsr_hg38_refs plink files
# GWAS input: hg38 TSV (not munged sumstats)

The run_ctwas rule is parameterised by cell_type and gwas wildcards, producing one output RDS per combination.

Technical Requirements

Category Detail
Software cTWAS R package
Container twas.sif (all rules; contains cTWAS + FUSION dependencies)
Input FUSION .wgt.RDat weights + .pos files (11-TWAS-weights); GWAS hg38 TSVs (07-prep-GWAS); SMR 1000G BIM (10-SMR)
Output Per-cell-type × GWAS cTWAS .rds result files; HTML report

Resource Profile

Rule Threads RAM Walltime
create_ld_matrices 5 20 GB 3d
copy_fusion_weights 1 5 GB 30 min
run_ctwas 6 96 GB 3d
ctwas_report 1 10 GB 1h

Run Composition

Phase Jobs Dimensions
create_ld_matrices 1 Once
copy_fusion_weights 19 19 cell types
run_ctwas 114 19 cell types × 6 GWAS
Back to top