10 · Summary Mendelian Randomisation (SMR)

Overview

This pipeline runs Summary Mendelian Randomisation (SMR) with HEIDI (Heterogeneity in Dependent Instruments) testing to identify genes where eQTL evidence is consistent with a shared causal variant with GWAS risk loci. It converts TensorQTL nominal output into BESD format, formats GWAS summary statistics for SMR, runs SMR + HEIDI for each cell type × GWAS combination, and summarises results.

Workflow Logic

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

    A(get_smr_binary) --> D(create_besd)
    B[TensorQTL nominal parquets] --> C(cat_tensorqtl_nom_snps)
    C --> E(create_query)
    F[SuSiE gene meta] --> E
    G(cat_refs) --> E
    E --> D
    G --> H(format_gwas)
    I[GWAS hg38 TSV] --> H
    D --> J(smr)
    H --> J
    A --> J
    G --> J
    J --> K(smr_report)

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

  1. get_smr_binary: Downloads the SMR v1.4.0 Linux binary from the Yang Lab.
  2. cat_refs: Concatenates per-chromosome 1000G hg38 PLINK reference files (BED/BIM/FAM/FRQ) into genome-wide files for SMR’s LD reference.
  3. cat_tensorqtl_nom_snps: Combines per-chromosome TensorQTL nominal Parquet files for each cell type into a single flat TSV using smr_cat_tensorqtl_nom_snps.py.
  4. create_query: Formats the concatenated nominal eQTL data into an SMR query file, filtering to eGenes passing FDR ≤ 0.05, aligning SNP IDs with the reference panel, and adding sample size (N=134).
  5. create_besd: Converts the query file into binary SMR BESD format (.besd, .bim, .epi files) using smr --make-besd.
  6. format_gwas: Reformats hg38 GWAS summary statistics into SMR .ma format using smr_format_gwas.R, aligning allele frequencies with the 1000G reference.
  7. smr: Runs SMR + HEIDI for each cell type × GWAS combination, testing each significant eGene for colocalisation with GWAS risk loci (p_SMR threshold: 5×10⁻⁸).
  8. smr_report: Renders an RMarkdown HTML report summarising SMR results filtered by p_SMR < 0.05 and p_HEIDI > 0.01.

SMR Filtering Criteria

Results are interpreted using the following thresholds (set in config/config.yaml):

p_smr: 0.05        # SMR association p-value threshold
p_heidi: 0.01      # HEIDI heterogeneity test threshold (reject if p < 0.01)
smr_window: 500    # LD window for HEIDI test (kb)
smr_gene: 'ENSG00000214435'  # Example gene (AS3MT) for QC plots

A gene passes SMR colocalisation if: p_SMR < 0.05 (association) AND p_HEIDI ≥ 0.01 (consistent with shared causal variant, not linkage).

Note

LD reference for SMR

SMR uses the 1000 Genomes hg38 PLINK files (shared with the S-LDSR pipeline) as its LD reference panel. These are concatenated from per-chromosome files by the cat_refs rule.

Technical Requirements

Category Detail
Software SMR v1.4.0
Container r_eqtl.sif (create_query, format_gwas, report); seurat5f.sif (create_besd)
Env modules compiler/gnu/5/5.0 (SMR binary requirements)
Input TensorQTL nominal Parquets (05-tensorqtl); GWAS hg38 TSVs (07-prep-GWAS); SuSiE gene meta (08-susie)
Output Per-cell-type × GWAS .smr result files; HTML report

Resource Profile

Rule Threads RAM Walltime
cat_tensorqtl_nom_snps 1 10 GB 1h
create_query 4 20 GB 1h
create_besd 4 20 GB 1h
format_gwas 4 20 GB 1h
smr 4 20 GB 1h
smr_report 4 20 GB 1h

Run Composition

Phase Jobs Dimensions
cat_tensorqtl_nom_snps 19 19 cell types
format_gwas 6 6 GWAS traits
smr 114 19 cell types × 6 GWAS
Back to top