Aims This study aims to identify predictive plasma protein biomarkers for anti‐tuberculosis drug‐induced liver injury (ATB‐DILI) and develop machine learning models for pre‐treatment risk stratification. Methods In this retrospective nested case–control study, proteomic profiling of pre‐treatment plasma from 24 patients (12 ATB‐DILI, 12 controls) identified differentially expressed proteins, which were validated by ELISA in an independent cohort (35 ATB‐DILI, 37 controls). Multiple machine learning algorithms were implemented to develop clinical prediction models and evaluate the prognostic value of the identified protein biomarkers. Results Proteomic analysis of pre‐treatment samples from the exploratory cohort identified five significantly differentially expressed proteins: antithrombin III, apolipoprotein D, carboxypeptidase B2, Chromogranin‐A, and Retinol‐binding protein 4. These proteins are functionally implicated in inflammatory responses, oxidative stress, and drug metabolism pathways. Validation using baseline plasma from an independent cohort confirmed consistent expression patterns for all five proteins ( p < 0.01), with directional changes matching the discovery phase findings. The random forest model, built on these pre‐treatment biomarkers, demonstrated robust predictive performance in the test set (AUC = 0.94, sensitivity = 90.0%, specificity = 90.0%, accuracy = 0.90). Importantly, consensus across multiple machine learning approaches (GBDT, SVM, GBM, etc.) confirmed predictive stability and generalizability of this protein signature (inter‐model AUC range: 0.85–0.96). Conclusion This study has successfully identified five pre‐treatment plasma protein signature that, when incorporated into machine learning models, may enable the prediction of ATB‐DILI risk, offering potential for early intervention in tuberculosis therapy.
Wang et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: