Abstract
The rising incidence of antimalarial drug resistance necessitates the urgent development of novel therapeutic agents targeting Plasmodium falciparum (Pf). This work presents a thorough computational framework that combines reinforcement learning (RL) driven molecular generation with multiscale validation techniques to find strong inhibitors of Pfdihydroorotate dehydrogenase (PfDHODH), a proven antimalarial target in the pyrimidine biosynthesis pathway. To create new molecular entities featuring DSM cores such as triazolopyrimidine and isoxazolopyrimidine scaffolds, we used REINVENT4.0 and its unique LibINVENT model. The 20,000 initial compounds produced by the RL approach were then further screened using filtering protocols that included drug-likeness parameters, such as synthetic accessibility scores, logP values, and quantitative estimation of drug-likeness (QED) values. Further screening by induced fit molecular docking studies against the PfDHODH (PDB ID:4RX0), identified 50 compounds exhibiting higher binding affinity than DSM265, a clinical candidate that failed in Phase II-a trials. Further, the top three hits were subjected to 100 ns molecular dynamics (MD) simulations using GROMACS with the CHARMM36 force field. Among them, Molecule 3 emerged as the lead candidate, demonstrating optimal stability based on RMSD, RMSF, radius of gyration, and total energy profiles. Further validation of binding interactions was achieved through the Quantum Theory of Atoms in Molecules (QTAIM) analysis, which offered detailed insights via bond critical point (BCP) evaluation. This integrated approach successfully identified novel PfDHODH inhibitors with superior theoretical binding properties compared to existing clinical candidates, demonstrating the probability of RL-guided molecular design in antimalarial drug discovery. The validated computational framework offers a robust platform for accelerating the identification of next-generation antimalarial therapeutics targeting drug-resistant Pf strains.