Robust Diagnostic and Therapeutic Biomarkers for Tuberculosis Identified Through Multi‐Omics and Mendelian Randomization Analysis

Tuberculosis (TB) remains a major global health challenge. In this study, we applied UPLC-MS/MS lipidomics and data-independent acquisition proteomics to profile plasma from healthy controls, active TB patients, and cured individuals to identify differentially expressed lipids and proteins. Mendelian randomization prioritized phosphatidylcholine (PC) lipids (PC(18:2/18:2), PC(14:0/20:4) and PC(18:0/20:4)) and proteins (haptoglobin [HP], retinol binding protein 4 [RBP4], coagulation factor XIII B subunit [F13B] and inter-alpha-trypsin inhibitor heavy chain 1 [ITIH1]) as candidate diagnostic and cure biomarkers. Binary multi-omics random-forest classifiers constructed with these markers achieved strong diagnostic (AUC = 0.967, 95% CI: 0.928-1.000) and cure-monitoring (AUC = 0.981, 95% CI: 0.956-1.000) performance, which was further assessed with ten-fold cross-validation. Integration with transcriptomic data and lipid-related gene analysis provided additional molecular support for HP. Independent validation in the GSE34608 cohort (AUC = 0.965) and ELISA verification (AUC = 0.969) confirmed HP's diagnostic utility at gene and protein levels. GSVA enrichment implicated HP in iron homeostasis and immune response pathways, suggesting a role in Mycobacterium tuberculosis infection and immune evasion through modulation of host iron metabolism. Overall, we present a robust lipid-protein biomarker panel and accurate multi-omics models for TB diagnosis and monitoring of cure, and propose HP as a promising biomarker and potential therapeutic target. These tools may improve clinical management and treatment evaluation.