Complex structural variations functionally inactivate the telomerase chaperone TCAB1 in osteosarcoma.

Tuberculosis (TB) remains a leading cause of death from a single infectious agent worldwide, with drug-resistant TB (DR-TB) representing a persistent threat to global health progress []. According to the World Health Organization (WHO) Global Tuberculosis Report 2025, an estimated 390,000 people developed multidrug-resistant or rifampicin-resistant TB (MDR/RR-TB) in 2024, accounting for 3.2% of new TB cases and 16% of previously treated cases []. Alarmingly, MDR/RR-TB caused approximately 150,000 deaths globally, with treatment success rates remaining suboptimal at 68–71% despite advances in all-oral regimens [, ]. A recent systematic review estimated the global prevalence of MDR-TB at 11.6% among tested populations, highlighting the substantial burden of resistance []. China continues to bear a substantial burden of DR-TB, accounting for 7.3% of global MDR/RR-TB cases in 2023 []. Although China has achieved notable progress in TB control—transitioning to a medium-low incidence country with an estimated 696,000 new TB cases in 2024, DR-TB remains a critical concern []. The estimated number of MDR/RR-TB cases in China was approximately 29,000 in 2023–2024, with treatment success rates of around 66% []. The long treatment duration (18–20 months), high pill burden, and severe adverse effects contribute to poor adherence and unfavorable outcomes []. Early identification of patients at high risk for unfavorable treatment outcomes is essential for implementing targeted interventions and improving prognosis [, ]. Traditional statistical methods have identified several predictors including age, nutritional status, and comorbidities [, ]. LASSO regression has been increasingly applied in TB research for variable selection and model development, including prediction of treatment outcomes in MDR/RR-TB populations [, ]. However, few studies have integrated machine learning-based approaches with interpretability analysis to develop clinically applicable prediction models specifically for DR-TB treatment outcomes. The SHapley Additive exPlanations (SHAP) framework offers advantages in elucidating individual-level predictions, enhancing clinical utility [].