Network-guided transcriptomics identifies drug-linked host-response biomarkers in tuberculosis disease-state transition.

Distinguishing active tuberculosis (ATB) from the post-therapy state remains a key challenge in disease monitoring, as host-response signatures are often broad and difficult to interpret. Robust and biologically interpretable markers are needed to capture this transition consistently across datasets. Here, we define a focused transcriptional signature associated with the shift from active disease (ATB_pre) to the post-therapy state (ATB_12m) using an integrative analysis of blood gene expression data. Differential expression analysis in the discovery cohort (n = 57) identified 1086 genes, which were reduced to a compact panel through network-guided prioritisation. External validation in an independent dataset identified 320 overlapping genes, with 95.9% showing consistent direction of change; this pattern was also retained in reverse-cohort sensitivity analysis. The final panel was dominated by interferon-associated genes, with STAT1 and GBP1 emerging as central markers. These genes showed higher expression in active disease and reduced expression at 12 months, a pattern preserved across cohorts. Functional analysis placed the signature within interferon signalling, innate immune regulation, and NOD-like receptor pathways. Internal machine learning assessment showed that the panel could distinguish ATB_pre from ATB_12m, with Random Forest and Support Vector Machine models achieving test-set ROC-AUC values of 0.969, supported by repeated fivefold cross-validation. Feature-level interpretation highlighted STAT1 and GBP1 as major contributors to classification. Drug-gene mapping provided exploratory pharmacological annotation of the panel using public drug-gene resources. The resulting eight-gene panel reflects transcriptional changes between active tuberculosis and the post-therapy state and may be useful for further evaluation in disease-state assessment.