DNA damage repair-related gene signatures delineate immune-distinct subtypes and diagnostic cues in tuberculosis

Background DNA damage repair-related genes (DDR-RGs) may shape tuberculosis (TB) pathogenesis, but their molecular characteristics, diagnostic utility, and cellular context remain insufficiently defined. Methods We integrated multiple GEO transcriptome datasets, identified differentially expressed genes, constructed weighted gene co-expression network analysis modules, and intersected these results with curated DDR-RGs to obtain candidate genes. An ensemble framework of 110 machine-learning model combinations was used to select diagnostic markers, followed by external validation, consensus clustering, immune infiltration analysis, single-cell RNA-seq evaluation, and Comparative Toxicogenomics Database-based compound-association analysis. Results The integrated training cohort included 134 samples; 94 differentially expressed genes and a 392-gene key module were identified, yielding 25 candidate DDR-RGs. The Ridge + SVM model performed best and selected ETV7, STAT1, FBXO6, and TAP1 as core DDR-RGs, achieving AUCs of 0.991, 0.925, 0.939, and 0.981 in the training and three validation cohorts. Consensus clustering separated TB patients into two DDR-related subtypes, with Cluster 1 enriched in immune-related pathways and characterized by distinct immune-cell infiltration. Single-cell analysis revealed cell-type-specific expression of core DDR-RGs in pulmonary TB, and compound-association analysis identified 35 candidate compounds, including six associated with multiple core genes. Conclusion These findings characterize DDR-related molecular heterogeneity in TB and support ETV7, STAT1, FBXO6, and TAP1 as candidate diagnostic biomarkers and exploratory candidates for host-directed studies.