Computational drug screening for drug-resistant tuberculosis: advances, challenges, and future perspectives

Tuberculosis (TB), caused by Mycobacterium tuberculosis, remains a major global health challenge, further exacerbated by the rapid emergence of drug-resistant strains. Despite continuous efforts, the discovery of novel anti-TB agents has lagged behind the evolution of resistance, largely due to the limitations of conventional experimental pipelines and the growing biological complexity of the pathogen. In recent years, computational drug discovery approaches-including virtual screening, artificial intelligence (AI)-driven modeling, and hybrid computational frameworks-have substantially expanded the chemical space accessible for early-stage anti-TB research. However, their translational success remains limited. This discrepancy reflects a fundamental mismatch between target-centric computational paradigms and the multi-scale biological constraints of Mtb, including permeability barriers, efflux mechanisms, metabolic state heterogeneity, lesion-level pharmacokinetics, and regimen-dependent therapeutic effects. This review critically evaluates current computational strategies for anti-TB drug screening, with particular emphasis on their performance under TB-specific biological and resistance constraints. We further explore computational approaches designed to address drug-resistant Mtb, including resistance-aware compound design, host-directed therapy, and combination regimen modeling. Finally, we explore key limitations in current computational pipelines, including challenges in data quality, biological integration, and regimen-level modeling, and propose a shift toward biologically grounded, regimen-aware, and systems-level frameworks. Bridging the gap between in silico prediction and clinical efficacy will require integrating multi-scale biological realism into computational pipelines, enabling more reliable translation of computational advances into effective anti-TB therapeutics.