EtienneNtumba/PRSice-Tanzania

PRSice-based polygenic risk score analysis and report for the Tanzania cohort. # PRSice Analysis: Fetal Hemoglobin Prediction in Tanzanian Sickle Cell Disease --- ## 📊 Overview This repository contains a comprehensive **polygenic risk score (PRS) analysis** with **p-value threshold optimization** for predicting fetal hemoglobin (HbF) levels in Tanzanian sickle cell disease (SCD) patients using **PRSice-2**. ### Study Highlights - 🧬 **10,000+ p-value thresholds** tested ($5 \times 10^{-8}$ to 0.5) - 👥 **1,527 individuals** from Tanzania (independent validation cohort) - 📈 **2.2 million SNPs** from genome-wide genotyping - 🎯 **Optimal threshold:** $p_T = 0.444$ (968,114 SNPs) - 📉 **Performance:** R² = 1.13% (minimal predictive power) - 🌍 **African population** - demonstrating PRS challenges in diverse populations - 📝 **Transparent reporting** - important negative result ### Key Finding Despite rigorous optimization, **genome-wide PRS explains only 1.13% of HbF variance** - demonstrating that **not all complex traits are suitable for PRS-based prediction**. This poor performance reflects HbF's **oligogenic architecture** (dominated by BCL11A) rather than the distributed polygenic effects required for successful PRS. --- ## 📑 Table of Contents - Background - Study Design - Methods - Key Results - Why PRS Performs Poorly - Clinical Implications - Repository Structure - Requirements - Installation - Usage - Interpretation - Citation - Team - License --- ## Background ### Polygenic Risk Scores (PRS) **Polygenic risk scores** aggregate genetic effects across many variants to predict trait values or disease risk: $$\text{PRS}_i = \sum_{j=1}^{M} w_j \cdot G_{ij}$$ where: - $w_j$ = effect size (weight) for variant $j$ from GWAS - $G_{ij}$ = genotype (0, 1, 2) for individual $i$ at variant $j$ - $M$ = number of variants included **Key Challenge:** Which variants to include? Which p-value threshold? ### P-Value Threshold Optimization PRSice solves this by testing thousands of thresholds: 1. **Strict** ($p_T = 5 \times 10^{-8}$): On …