Deep Learning-Driven Discovery of FDA-Approved BCL2 Inhibitors: In Silico Analysis Using a Deep Generative Model NeuralPlexer for Drug Repurposing in Cancer Treatment

ABSTRACT Finding strong inhibitors of the BCL2 target, which is essential for controlling apoptosis and ensuring the survival of cancer cells, has prompted research into FDA-approved drugs. This study uses an advanced deep generative model called NeuralPlexer to produce protein-ligand complex conformations one-by-one and perform in silico analysis. This is the first time in the literature using NeuralPlexer in virtual screening, as we comprehensively evaluate the conformations of an FDA-approved drug library to ascertain their potential efficacy in suppressing BCL2 by utilizing NeuralPlexer’s capabilities. Obtained results were re-confirmed by physics-based molecular simulations and neural relational inference (NRI) analysis. Our study reveals several intriguing candidates such as Lathyrol and Fadrozole with potent inhibitory interactions with the BCL2 target, offering important new information for repurposing currently available drugs in cancer treatment. This work highlights the promise of deep learning technology in pharmaceutical research by integrating NeuralPlexer into the process of drug development, while also improving the accuracy of predictions made about protein-ligand interactions.