From Mycobacterial Persistence to Novel Therapeutics
TB, caused by Mycobacterium TB (Mtb), is a leading infectious killer, with rising drug resistance demanding novel targets. The bacterial toxin-antitoxin system regulates stress responses. The MenT3 toxin inhibits protein synthesis by adding CMP to transfer ribonucleic acid (tRNA), promoting bacterial persistence within hosts. While removing MenT3 weakens Mtb, no inhibitors exist.
Previous computational studies have successfully combined artificial intelligence (AI) with physics-based methods for other targets, but this approach has not been applied to MenT3. To fill this gap, the present study integrated generative machine learning with pharmacokinetic filtering, molecular docking, dynamics simulations, and density functional theory (DFT) to identify novel MenT3 inhibitors.
An Integrated AI and Physics-Based Virtual Screening Pipeline
This study employed a hybrid computational pipeline integrating machine learning and physics-based methods to identify novel inhibitors of the MenT3 toxin from Mycobacterium tuberculosis. All computations were performed using SilicoXplore, a cloud-based drug discovery platform that combines open-source and proprietary tools, including AutoDock Vina, GROMACS, ADMET-AI, and PharmacoNet.
The workflow began with generative AI using REINVENT4's de novo design and reinforcement learning to create approximately 100,000 novel molecules. These were first filtered through ADMET-AI for pharmacokinetic properties, including drug-likeness, toxicity, and bioavailability. Retained molecules underwent pharmacophore screening using PharmacoNet to identify compounds with key binding features matching the MenT3 active site.
For molecular docking, the MenT3 crystal structure containing the co-crystal ligand cytidine triphosphate (CTP) was prepared by removing water molecules, adding hydrogen atoms, and assigning Gasteiger charges. Docking was performed using AutoDock Vina with grid coordinates centered on the CTP binding site. The protocol was validated by self-docking CTP, achieving root-mean-square deviation (RMSD) < 2 angstrom (Å). Following docking, a similarity search using Tanimoto coefficients identified molecules structurally similar to CTP.
Compounds containing a pyrimidine ring were subjected to 20 nanoseconds (ns) MD simulations using GROMACS with the CHARMM36 force field and TIP3P water model. The top five candidates and the MenT3-CTP complex underwent extended 100 ns simulations. Binding free energies were calculated using molecular mechanics generalized born surface area (MM-GBSA) from simulation trajectories.
Finally, DFT using PSI4 with the B3LYP functional and 6-31G(d) basis set evaluated electronic properties, including the highest occupied molecular orbital and lowest unoccupied molecular orbital (HOMO-LUMO) gaps, molecular electrostatic potentials, and global reactivity descriptors.
Machine Learning-Driven Discovery and Physics-Based Validation
This study successfully identified five novel MenT3 toxin inhibitors using a hybrid computational pipeline. Starting with 100,000 de novo-generated molecules from REINVENT4, pharmacokinetic filtering with ADMET-AI retained 11,625 compounds. Pharmacophore screening using PharmacoNet further narrowed these to 1,724 molecules, followed by triple-replicate molecular docking against the MenT3 crystal structure.
Using the average binding energy of the co-crystal ligand CTP as a threshold, 1,481 molecules with higher affinity were selected. Similarity searching (Tanimoto coefficient ≥0.5) combined with the pyrimidine ring requirement yielded 14 candidates.
Initial 20 ns MD simulations evaluated stability parameters including RMSD, root-mean-square fluctuation (RMSF), radius of gyration, hydrogen bonds, and solvent-accessible surface area. MM-GBSA binding free energy calculations identified five molecules with superior or comparable binding affinities relative to CTP. Extended 100 ns MD simulations confirmed stable binding, with MenT3_M2 showing the strongest binding free energy. Binding interaction analysis revealed key hydrophobic contacts and a hydrogen bond similar to CTP.
DFT calculations demonstrated HOMO-LUMO energy gaps exceeding six electronvolts (eV) for all candidates, indicating favorable stability. Molecular electrostatic potential maps identified electron-rich regions (carbonyl oxygens, nitrogen heteroatoms) critical for binding.
Download the PDF of this page here
Reactivity descriptors showed MenT3_M3 exhibited electrophilic character most similar to CTP. While these computational findings are promising, the authors note limitations, including a lack of experimental validation, and recommend future in vitro and in vivo studies to confirm anti-tubercular activity.
From In Silico Discovery to Experimental Validation
This study successfully identified five novel MenT3 inhibitors using a hybrid AI and physics-based pipeline. MenT3_M2, MenT3_M4, and MenT3_M5 exhibited superior binding free energies compared to the co-crystal ligand CTP. DFT analysis confirmed favorable electronic properties, with HOMO-LUMO gaps (6.75-7.60 eV) similar to CTP (7.88 eV), indicating stability and reactivity suitable for MenT3 inhibition. While these computational findings are promising, thorough experimental validation through in vitro and in vivo studies is essential before determining their therapeutic potential against tuberculosis.
Journal Reference
Alsarra et al. (2026). Identification of potential MenT3 inhibitors for Mycobacterium tuberculosis using the generative artificial intelligence and SilicoXplore platform. Scientific Reports. DOI:10.1038/s41598-026-49174-y, https://www.nature.com/articles/s41598-026-49174-y
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.