Multiomics and Machine Learning Identify Immunometabolic Biomarkers for Active Tuberculosis Diagnosis Against Nontuberculous Mycobacteria and Latent Tuberculosis Infection

Abstract

This study utilized multiomics combined with a comprehensive machine learning-based predictive modeling approach to identify, validate, and prioritize circulating immunometabolic biomarkers in distinguishing tuberculosis (TB) from nontuberculous mycobacteria (NTM) infections, latent tuberculosis infection (LTBI), and other lung diseases (ODx). Functional omics data were collected from two discovery cohorts (76 patients in the TB-NTM cohort and 72 patients in the TB-LTBI-ODx cohort) and one validation cohort (68 TB patients and 30 LTBI patients). Mutiomics integrative analysis identified three plasma multiome biosignatures that could distinguish active TB from non-TB with promising performance, achieving area under the receiver operating characteristic curve (AUC) values of 0.70-0.90 across groups in both the discovery and validation cohorts. The lipid PC(14:0_22:6) emerged as the most important predictor for differentiating active TB from non-TB controls, consistently presenting at lower levels in the active TB group compared with its counterparts. Further validation using two independent external data sets demonstrated AUCs of 0.77-1.00, confirming the biomarkers' efficacy in distinguishing active TB from other non-TB groups. Our investigation highlights lipids as promising biomarkers for classifying TB, NTM, LTBI, and ODx. Rigorous validation further indicates PC(14:0_22:6) as a TB differential diagnostic biomarker candidate.

Keywords: biomarkers; diagnosis; machine learning; multiomics; nontuberculous mycobacteria infection; tuberculosis.