Predicting Long COVID Persistence Using Mitochondrial Gene Expression Profiles: An XGBoost-Based Machine Learning Approach with External Validation

This repository contains the complete analysis pipeline, preprocessed datasets, and trained machine learning models supporting the manuscript: "Predicting Long COVID Persistence Using Mitochondrial Gene Expression Profiles: An XGBoost-Based Machine Learning Approach with External Validation" Study Overview: We developed an XGBoost-based machine learning classifier using 149 mitochondrial gene expression profiles (MitoCarta3.0) from peripheral blood transcriptomes of 343 individuals across 6 independent GEO cohorts (GSE267625, GSE169687, GSE152418, GSE222253, GSE265753, GSE235938). The model distinguishes Long COVID persistence from recovery with high accuracy (nested CV AUC = 0.999 ± 0.001; external validation AUC = 0.984, 95% CI: 0.959–1.000). Repository Contents: longcovid_ML_FULL_RESULTS.zip — Complete analysis package including Python scripts, trained XGBoost model, cross-validation results, feature importance rankings, and publication-quality figures (TIFF, 300 DPI). X_full_transcriptome_log2cpm.csv.gz — Full transcriptome expression matrix (18,335 genes × 343 samples, log2-CPM normalized, ComBat batch-corrected). X_mito_genes_log2cpm.csv.gz — Mitochondrial gene expression matrix (149 MitoCarta3.0 genes × 343 samples, log2-CPM normalized, ComBat batch-corrected). README.md — Detailed documentation including data dictionary, reproducibility instructions, and citation information.