PRS Pipeline

A modular Python pipeline for Polygenic Risk Score (PRS) calculation using pgsc_calc, with statistical validation and clean score export.

---

Pipeline Overview

Step 1: Preprocessing    → detect input files, auto-generate samplesheet
Step 2: Run pgsc_calc    → run Nextflow + pgsc_calc
Step 3: Validate         → AUC, R², Odds Ratios, Cohen's d, plots
Step 4: Export Scores    → clean CSV/XLSX table ready to join to any metadata

---

Prerequisites

1. Python dependencies

pip install -r requirements.txt

2. Nextflow

# Requires Java 11+
curl -s https://get.nextflow.io | bash
mv nextflow ~/bin/   # or any directory on your PATH

Full instructions: https://www.nextflow.io/docs/latest/install.html

3. pgsc_calc

pgsc_calc runs via Nextflow and does not require a separate installation. It will be pulled automatically the first time you run Step 2.

Supported execution profiles (set in config.yaml):

• conda — recommended for HPC/local use
• docker — requires Docker daemon
• singularity — recommended for clusters without Docker

Full pgsc_calc documentation: https://pgsc-calc.readthedocs.io

---

Setup

git clone https://github.com/jordanlab/PRS.git
cd PRS
pip install -r requirements.txt

Edit config.yaml with your paths before running any step.

---

Configuration (config.yaml)

# Path prefix to genotype files (glob-style, e.g. /path/to/ukb22418_c*_b0_v2)
genotype_prefix: "/path/to/your/genotypes/prefix"

# Where to write the auto-generated samplesheet
samplesheet_output: "/path/to/samplesheet.csv"

# pgsc_calc settings
output_dir:    "/path/to/pgsc_results/"
target_build:  "GRCh37"   # or GRCh38
profile:       "conda"    # docker, conda, or singularity

# Validation settings (Step 3)
data_file:      "/path/to/merged_prs_phenotype.csv"
val_output_dir: "/path/to/val_output/"

# Score export settings (Step 4)
scores_file:    ""   # leave blank to auto-detect from output_dir
export_output:  "/path/to/prs_scores_export.xlsx"

---

Usage

Run each step individually:

python main.py --step preprocess
python main.py --step run_pgsc_calc
python main.py --step validate
python main.py --step export_scores

To use a different config file:

python main.py --step preprocess --config /path/to/my_config.yaml

---

Step-by-Step Guide

Step 1 — Preprocessing

Detects your genotype files and auto-generates a pgsc_calc samplesheet.

Supports:

• PLINK 1 binary (bfile: .bed/.bim/.fam)
• PLINK 2 binary (pfile: .pgen/.pvar/.psam)
• VCF / VCF.gz

Handles both:

• Per-chromosome files (e.g. ukb22418_c1_b0_v2, ukb22418_c2_b0_v2 …)
• Single merged files

After auto-generation, you will be prompted to review the samplesheet in your terminal editor before proceeding.

---

Step 2 — Run pgsc_calc

Builds and runs the Nextflow command. You will be prompted for:

• PGS Catalog ID (e.g. PGS000039) — pgsc_calc downloads the scoring file automatically
• Samplesheet path (from Step 1)
• Optional flags: --min_overlap, --ancestry_method, --ref, --liftover, --keep_ambiguous

The full command is printed for your review before execution.

pgsc_calc output will be in output_dir as set in config.yaml. Key files:

• score/*/aggregated_scores.txt.gz — PRS scores per sample
• HTML report — interactive QC report

---

Step 3 — Validation

Requires a merged file with PRS scores, case/control status, and covariates (age, sex, PCs, etc.). You will be prompted to identify the relevant columns.

Outputs:

• validation_summary.csv — AUC, incremental R², Cohen's d, t-test p-value
• odds_ratios.csv — OR with 95% CI for all model terms
• roc_curve.png
• risk_gradient.png — case prevalence per PRS decile
• prs_distribution.png — overlapping PRS density for cases vs controls

---

Step 4 — Export Scores

Parses aggregated_scores.txt.gz from pgsc_calc and exports a clean, joinable table.

Output: prs_scores_export.csv and prs_scores_export.xlsx

Columns: Patient_ID, PRS (or per-ancestry score columns if multiple are present).

To merge with your own patient metadata:

import pandas as pd

meta   = pd.read_excel("patient_metadata.xlsx")
prs    = pd.read_csv("prs_scores_export.csv")
merged = meta.merge(prs, on="Patient_ID", how="left")
merged.to_excel("patient_metadata_with_PRS.xlsx", index=False)

---

Input Format Notes

pgsc_calc Samplesheet

Auto-generated by Step 1. Format follows the pgsc_calc samplesheet specification.

Column	Description
sampleset	Name for your cohort
vcf_path	Absolute path to VCF/VCF.gz (if applicable)
bfile_path	Prefix for PLINK bfile (if applicable)
pfile_path	Prefix for PLINK pfile (if applicable)
chrom	Chromosome number (leave blank for merged files)
format	vcf, bfile, or pfile

---

Maintained by

Valli Sree Lasya Pasumarthy
MS Bioinformatics, Jordan Lab
Georgia Institute of Technology