sukhjot-spec/tb_amr_variant_calling

WGS variant-calling and explainable ML pipeline identifying and characterizing compensatory mutations linked to MDR-TB in African M. tuberculosis isolates - combining Fisher's exact testing, SHAP/EBM explainability, phylogenetic conservation, and protein structural analysis into one integrated evidence table. # M. tuberculosis WGS Variant Calling & Compensatory Mutation Analysis Pipeline ### Comparative Genomic and Explainable Machine Learning Analysis of Compensatory Mutations Associated with Multidrug Resistance in African Mycobacterium tuberculosis **Project status: COMPLETE.** All three research objectives, all eleven analytical steps, and a final publication-figures step have been executed, verified against real data at every stage, and (where problems were found along the way) fixed and re-verified. This README documents the full pipeline from raw sequencing reads through to the final, evidence-integrated candidate table and publication figures. --- ## Table of Contents 1. Project Overview 2. Research Objectives 3. Dataset 4. Repository Structure 5. Environment Setup 6. Pipeline Architecture 7. Step-by-Step Pipeline Execution 8. Post-VCF Processing (Steps 1-3, Objective 1) 9. Explainable Machine Learning (Steps 4-7, Objective 2) 10. Evolutionary, Lineage, and Structural Analysis (Steps 8-10, Objective 3) 11. Final Integration and Publication Figures (Steps 11-12) 12. Output Files and Their Role 13. Tools and Software 14. Key Findings Summary 15. Known Issues Found and Fixed During This Project 16. Progress Status 17. Citation --- ## 1. Project Overview This repository contains the complete computational pipeline for whole-genome sequencing (WGS) based variant calling, antimicrobial resistance (AMR) analysis, explainable machine learning, and multi-layer evolutionary/structural characterisation of candidate compensatory mutations in *Mycobacterium tuberculosis* clinical isolates. The pipeline processes raw sequencing data from NCBI's Sequence Read Archive (SRA), aligns reads to the H37Rv reference genome, calls variants, identifies statistically significant compensatory-mutation candidates, trains and explains four machine learning models for MDR prediction, and characterises the surviving candidates' lineage distribution, evolutionary conservation, and prot …