Sex-specific gene expression signatures as biomarkers in predicting tuberculosis relapse

Background: Tuberculosis (TB) relapse after treatment completion remains a major challenge, with limited reliable biomarkers and important sex-dependent biological differences. Methods: Gene expression data from the PREDICT-TB cohort were analysed with a three-stage machine learning feature selection pipeline on the male subset. The final minimal gene panel was evaluated with Random Forest and XGBoost using AUC, sensitivity, and specificity via repeated k-fold cross-validation and external validation. Results: A 7-gene signature (ENTPD2, KANK1, ANAPC13, PI4K2A, INTS1, ZNF439 and NFATC2IP-AS1) discriminated relapsed from cured patients internally; RF yielded AUC = 0.9656. Performance declined on external cohorts, indicating limited generalizability. Conclusion: The signature is a strong mechanistic hypothesis for male-specific relapse susceptibility, but cross-cohort harmonization and a female-specific signature are needed for clinical translation.