Identifying mitochondria-related signatures for tuberculosis diagnosis through machine learning on single-cell...

Tuberculosis (TB) remains a major cause of infectious disease mortality. Early diagnosis is crucial for curbing transmission and initiating timely treatment. However, the lack of reliable non-sputum-based diagnostic tools often delays prompt detection. Since mitochondrial dysregulation facilitates Mycobacterium tuberculosis (MTB) evasion, we explored mitochondria-related gene signatures as diagnostic biomarkers. By integrating microarray (GSE19491) and single-cell RNA sequencing (scRNA-seq; SRP247583) data from active tuberculosis disease (TBD), latent tuberculosis infection (TBI), and healthy controls (HC) obtained from the Gene Expression Omnibus (GEO) database, we identified ten mitochondria-related differentially expressed genes (MitoDEGs) -STAT2, CASP1, SCO2, PRELID1, COX7B, COX6A1, TSPO, IFI6, ATG3, and COX7A2- in the monocytic lineage. Enrichment analysis revealed that these ten MitoDEGs were primarily enriched in oxidative phosphorylation. Quantitative PCR (qPCR) validated the upregulation of these genes in an H37Rv-infected THP-1 cell model (P < 0.01). Using the Least Absolute Shrinkage and Selection Operator (LASSO) and Support Vector Machine-Recursive Feature Elimination (SVM-RFE) algorithms, we prioritized three hub markers (STAT2, CASP1 and COX7B) to construct a blood-based diagnostic model. The SVM-based model achieved robust diagnostic performance in differentiating TBD in the training (n = 125; AUC = 0.852), validation (n = 30; AUC = 0.885), and two testing sets: GSE54992 (n = 15; AUC = 0.907) and GSE34608 (n = 26; AUC = 0.993). Moreover, an independent clinical cohort further confirmed its efficacy (n = 52; AUC = 0.909) in discriminating TBD from non-TB controls. In summary, we developed a three-MitoDEG model that shows promising diagnostic performance for TBD and was preliminarily validated, offering a scalable, non-sputum alternative for triage in resource-limited settings.