Machine Learning‐Based Discovery of Natural PknB Inhibitors Against Drug‐Resistant Mycobacterium tuberculosis With Multi‐Scale Modeling and Quantum Mechanical Validation
TL;DR
An integrated computational workflow to identify natural small molecules with high potential to bind protein kinase B (PknB) identifies promising PknB inhibitors for further experimental validation and offers an efficient platform to accelerate natural‐product‐based anti‐TB drug discovery.
Abstract
Tuberculosis (TB) caused by Mycobacterium tuberculosis (Mtb) remains a major global health threat, particularly with rising drug resistance. Protein kinase B (PknB), an essential mycobacterial Ser/Thr kinase absent in humans, is a promising therapeutic target. This study describes the use of an integrated computational workflow to identify natural small molecules with high potential to bind PknB. Structure‐based virtual screening, machine‐learning algorithms, and deep‐learning bioactivity prediction identified six compounds with high predicted pIC 50 values. The AI‐based ADMET assessment showed promising pharmacokinetic and toxicity profiles, and the redocking and residue‐interaction analyses suggested strong binding affinities and interactions. The all‐atom molecular dynamics simulations showed the stability of the protein–ligand complexes over 1000 ns. CNP0362879, CNP0343061, and CNP0413118 were identified as the most favorable binders by MM/GBSA free‐energy calculations. DFT and QM/MM analyses also characterized the electronic properties related to the molecular reactivity and binding. Network pharmacology linked the prioritized compounds with therapeutically relevant targets and pathways. This AI‐integrated multiscale approach offers an efficient platform to accelerate natural‐product‐based anti‐TB drug discovery and identifies promising PknB inhibitors for further experimental validation.