Identification of Gene Expression Biomarkers Predictive of Latent Tuberculosis Infection Using Machine Learning Approaches

Abstract

Latent tuberculosis infection (LTBi) affects nearly a quarter of the global population, yet current diagnostic methods are limited by low sensitivity and specificity. This study applied an integrative bioinformatics framework, incorporating machine learning techniques, to identify robust gene expression biomarkers associated with LTBi. We analyzed four publicly available transcriptomic datasets from peripheral blood mononuclear cells (PBMCs), representing latent, active, and healthy states. Differentially expressed genes (DEGs) were identified, followed by gene ontology (GO) enrichment, functional clustering, and miRNA interaction analysis. Semantic similarity, unsupervised clustering, and pathway enrichment were applied to refine the gene list. Key biomarkers were prioritized using receiver operating characteristic (ROC) curve analysis, with CCL2 and CXCL10 emerging as top candidates (AUC > 0.85). This multi-step approach demonstrates the potential of combining transcriptomic profiling with established machine learning and bioinformatics tools to uncover candidate biomarkers for improved LTBi detection, and it also provides a foundation for future experimental validation.