gwas-pipeline

Python package for packaging the full GWAS tutorial workflow behind a consistent CLI and reusable step modules.

It installs a pure GWAS package with packaged step1 through step9 modules, while keeping the legacy standalone step scripts as thin compatibility wrappers.

Scope of the first version

This scaffold focuses on the minimum post-GWAS path:

• summary statistics loading
• configurable source-to-standard column mapping
• variant harmonization
• colocalization wrapper boundary
• Mendelian randomization wrapper boundary
• batch workflow orchestration by gene and cell type

The package is designed to keep analysis logic in Python while calling specialized external tools from adapters when needed.

R dependencies

The real colocalization and MR wrappers expect:

• Rscript available on PATH
• R package coloc
• R package MendelianRandomization

Install

pip install -e .

GWAS tutorial CLI

After pip install -e ., you can run the packaged step modules through either entrypoint:

gwas-pipeline step3 --gwas /path/to/results.tsv --gff /path/to/genes.gff --output-dir /tmp/step3
python -m gwas_pipeline step8 --vcf /path/to/region.vcf.gz --region 22:1-100000 --class-file class.csv --output-dir /tmp/step8
gwas-pipeline doctor --profile plink_env

The original gwas_step*_project/*.py files remain available and now forward to the packaged modules in src/gwas_pipeline/steps/.

Runnable GWAS template

There is also a pure-Python GWAS teaching template that runs without plink, numpy, or pandas. It is intended for learning the workflow and for adapting to small tabular datasets before swapping in external tools for real production-scale GWAS.

Run the bundled demo:

python3 scripts/run_simple_gwas_template.py \
  --config examples/simple_gwas_template/config.json

The template writes:

• gwas_results.tsv
• candidate_hits.tsv
• summary.txt
• manhattan.svg
• qq_plot.svg

Real-data preprocessing

For split cell-type eQTL files similar to the Bryois-style layout, first build standardized tables:

python scripts/build_standardized_eqtl.py \
  --eqtl-root /home/y413109/project/eqtl \
  --snp-map /home/y413109/project/eqtl/snp_pos.txt \
  --output /home/y413109/project/eqtl/standardized/brain_eqtl_standardized.tsv.gz

python scripts/build_standardized_gwas.py \
  --gwas /home/y413109/project/eqtl/gwas/summary_stats_AD_mapped \
  --output /home/y413109/project/eqtl/standardized/ad_gwas_standardized.tsv.gz

Then point examples/run_real_postgwas.yaml at those standardized outputs.

Real input expectations

The package standardizes input columns into this internal schema:

• chrom
• position
• effect_allele
• other_allele
• beta
• se
• p_value
• gene_id for eQTL only
• cell_type for eQTL only
• maf optional today, reserved for future coloc improvements

Use inputs.eqtl.columns and inputs.gwas.columns in YAML to map your real source column names onto those standard names.