Pathway-Level Germline DDR Rare Variants in Metastatic Prostate Cancer — Analysis Code

This repository contains the analysis code and de-identified derived data supporting the manuscript:

Pathway-Level Germline Rare Variant Enrichment Across DNA Damage Response Genes in Metastatic versus Non-Metastatic Prostate Cancer: An Extreme Phenotype Study. Lin Y-Y et al., submitted to European Urology (2026). Preprint: https://doi.org/10.1101/2025.04.28.25326584 (v1 — an updated preprint matching this submission will be posted to medRxiv on acceptance.)

Most genetic studies of prostate cancer have addressed cancer initiation; this study targets a distinct phenotype of metastatic progression after diagnosis using an extreme-phenotype cohort design that sequences patients at opposite ends of the post-diagnosis course. The finding is a pathway-level germline rare-variant burden across DNA Damage Response (DDR) genes that distinguishes the two arms, with convergent contributions from multiple DDR genes rather than a single dominant variant. The pathway-level signal replicates in an independent Australian extreme-phenotype cohort and at the pathway level in the Pan Prostate Cancer Group (n = 976). In vitro functional studies of variants currently classified as uncertain or benign in clinical databases reveal measurable phenotypic effects, pointing to patients who may benefit from DDR-targeted therapy but are not flagged by current variant-classification criteria.

A persistent DOI for the code and derived data is minted via Zenodo and listed in the manuscript's Data Availability statement.

Repository contents

In this repo

• All scripts used to generate the statistical results, tables, and figures reported in the manuscript.
• The de-identified clinical cohort table (data/clinical/clinical.tsv, 52 patients) — the real Table 1 cohort, cleared for public release.
• Primary pipeline input tables (data/raw/) — per-patient and per-gene aggregate matrices (gnsRV burden, oncoplot matrices, cBioPortal exports, quantified wet-lab assay readings).
• A synthetic cohort (data/synthetic/) matching the schema of the real data, for end-to-end smoke testing without controlled-access credentials.
• De-identified summary tables (data/derived/): the prespecified DDR gene panel, the 56-variant genome-wide screen list (with gnomAD MAFs, FATHMM-MKL, EVEE, and ClinVar annotations), Australian EPC carrier counts, and PPCG gene-level association summary.
• A claim-to-script crossreference (docs/METHODS_CROSSREF.md) that maps each numerical result in the manuscript to the file that produces it.

Not in this repo

Raw FASTQ, BAM, VCF, and raw wet-lab image stacks are not in this repository. They remained controlled-access and are being deposited to a public controlled-access repository before manuscript publication (see docs/DATA_AVAILABILITY.md). The aggregate matrices and quantified readings that the analysis pipeline actually consumes are committed under data/clinical/ and data/raw/

A note on tools

Every stage uses standard, widely validated tools (e.g. BWA-MEM, Strelka2, ANNOVAR, Firth penalized regression, SKAT) so the findings do not depend on bespoke, unvalidated code and can be reproduced with any equivalent workflow. The methodological contribution is the extreme-phenotype cohort design and the rare-variant prioritization strategy, not new software.

Quickstart

Python 3.11 with pandas, numpy, scipy, matplotlib, openpyxl, marimo (for notebooks/), and pytest (for the smoke tests). Exact versions are pinned in environment.yml / requirements.txt. Nothing in code/ requires R: the PPCG gene-level SKAT tests are run by the PPCG consortium, not from this repo.

git clone https://github.com/yenyilin/rarevariant-mPC.git
cd rarevariant-mPC

# Option A — conda (canonical, matches environment.yml exactly)
conda env create -f environment.yml
conda activate rarevariant-mpc

# Option B — pip on python.org Python 3.11 (the setup the maintainer actually used)
python3.11 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Option C — uv (much faster than pip; same requirements.txt)
uv venv --python 3.11
source .venv/bin/activate
uv pip install -r requirements.txt

# 1. Generate the synthetic cohort
python data/synthetic/generate_synthetic_clinical.py

# 2. Build the analysis-ready table
python code/01_cohort_prep/prepare_epc_cohort.py --synthetic

# 3. Reproduce Firth regression (synthetic OR will not match published OR=26.80)
python code/04_firth_regression/ddr_logistic_regression.py

# 4. Run the test suite
pytest

# 5. Open an interactive walkthrough (marimo)
marimo edit notebooks/01_ddr_convergent_walkthrough.py

To reproduce the published numerical results, just re-run steps 2 onward without the --synthetic flag — the real cohort tables are already committed under data/clinical/ and data/raw/. See docs/REPRODUCIBILITY.md for the full command list.

Reproducing specific manuscript claims

See docs/METHODS_CROSSREF.md for the full table. Examples:

Manuscript claim	Script
DDR gnsRV enrichment, p = 4.57 × 10⁻⁶	code/03_ddr_enrichment/ddr_arm_compare.py
Firth OR = 26.80, 95% CI 3.07–3528.42	code/04_firth_regression/ddr_logistic_regression.py
Mann-Whitney follow-up comparison, p < 0.001	code/05_time_bias/ddr_followup_plot.py
Australian EPC binomial, p = 0.027–0.050	code/06_replication/australian_binomial.py

Repository layout

.
├── code/
│   ├── 01_cohort_prep/        Build analysis table from clinical + genotype inputs
│   ├── 02_variant_calling/    BWA / Strelka2 / annotation pipeline (documented)
│   ├── 03_ddr_enrichment/     DDR pathway-burden tests (Mann-Whitney, gene-set bootstrap, per-patient p)
│   ├── 04_firth_regression/   Firth penalized logistic regression (OR=26.80)
│   ├── 05_time_bias/          Mann-Whitney + Spearman time-bias analyses
│   ├── 06_replication/        Australian EPC binomial (PPCG SKAT done by consortium)
│   ├── 07_functional/         Quantification of in vitro assays
│   └── 08_figures/            One script per main and supplementary figure
├── data/
│   ├── clinical/              De-identified clinical cohort table (52 patients)
│   ├── raw/                   Primary pipeline inputs (per-patient aggregate matrices)
│   ├── derived/               Aggregate / de-identified summary tables
│   ├── synthetic/             Simulated cohort matching the real schema
│   └── processed/             Pipeline outputs (gitignored; regenerated at runtime)
├── docs/
│   ├── DATA_AVAILABILITY.md   Accession numbers and access procedure
│   ├── PRECOMMIT_CHECKLIST.md Pre-release checks (PHI / controlled-data scan)
│   └── METHODS_CROSSREF.md    Manuscript claim → script crossreference
├── notebooks/                 Reviewer-facing marimo walkthroughs (.py)
├── tests/                     Sanity checks runnable on synthetic data
└── .github/workflows/         CI (pytest on synthetic data)

Citation

@article{rarevariant_mpca_2026,
  title  = {Pathway-Level Germline Rare Variant Enrichment Across DNA
            Damage Response Genes in Metastatic versus Non-Metastatic
            Prostate Cancer: An Extreme Phenotype Study},
  author = {Lin, Y.-Y. and others},
  year   = {2026},
  note   = {Preprint: doi:10.1101/2025.04.28.25326584;
            Code and derived data: doi:10.5281/zenodo.20317985}
}

License

Code is released under the MIT License (LICENSE). Documentation and the public data tables (data/derived/, data/clinical/, data/synthetic/) are released under CC-BY-4.0. Raw sequencing data deposited externally remains under controlled access governed by the original consent.

Contact

Issues and questions: please open a GitHub issue. Data access: see docs/DATA_AVAILABILITY.md.