Multi-scale computational methods identify pivotal tuberculosis vulnerabilities and predict drug resistance with high fidelityComputational Methods Identify New Targets for Tuberculosis Treatment

Tuberculosis (TB), driven by Mycobacterium tuberculosis (Mtb), remains a leading cause of mortality from infectious disease worldwide.​ The global dissemination of multidrug-resistant (MDR) and extensively drug-resistant (XDR) strains—coupled with protracted treatment regimens and the persistence of latent infection—has not only compromised first-line chemotherapies but also underscored the stagnation of conventional drug development pipelines. In this review, we systematically synthesize how multi-scale computational strategies are reshaping anti-TB drug discovery and explicitly delineate their hierarchical integration—a cohesive framework that, to our knowledge, has yet to be comprehensively mapped in the existing literature.​ We survey four interconnected computational layers: (i) systems and network pharmacology for macroscopic target and pathway prioritization; (ii) machine learning (ML) for mesoscopic virtual screening, resistance classification, and ADMET prediction; (iii) molecular docking for microscopic elucidation of ligand–target binding modes; and (iv) molecular dynamics (MD) simulations for atomic-level validation of binding poses and assessment of complex stability. Representative studies demonstrate that this workflow has pinpointed pivotal TB vulnerabilities (e.g., MmpL3, InhA, DprE1, ATP synthase, and FtsZ), enabled ML models to predict drug resistance from genomic data with high fidelity, and identified promising candidates such as vazabitide A (predicted binding affinity of −85.8 kcal/mol against Pks13) alongside repurposed agents (e.g., aldoxorubicin, quarfloxin). Critically, these computational predictions are reinforced by orthogonal experimental validation spanning in vitro enzymatic and phenotypic assays, in vivo infection models, clinical pharmacokinetic profiling, and multi-omics corroboration. We conclude that the top-down pipeline-spanning network pharmacology to ML, docking, and MD-integrated with bottom-up feedback constitutes a closed-loop paradigm that accelerates lead identification while addressing emergent drug resistance.​ Nevertheless, inherent limitations in predictive reliability and cross-layer error propagation necessitate rigorous downstream experimental validation. Looking forward, generative artificial intelligence, self-driving laboratories, and standardized cross-method benchmarking are poised to consolidate this multi-scale pipeline into a robust engine for next-generation anti-TB therapeutics.