Convolutional Neural Network Analysis of Molecular Docking for Drug Discoveries

In this study, we performed a comparative study of series of convolutional neural networks (CNN) including search-based methods such as GNINA and a diffusion generative model (DiffDock) approach to optimize the drug design process for Mouse Double Minute 2 (MDM2) proteins and SARS-CoV-2 main protease. MDM2 inhibitors are of great interest to pharmacological research because MDM2 is a key protein involved in the downregulation of p53 in the presence of cellular damage. The protein p53 activates cell cycle arrest or in the presence of significant DNA or cell damage begins the signaling cascade for apoptosis. In excess, MDM2 prevents these processes from occurring when desired and allows for unregulated damaged DNA expression and cancer. Main protease of SARS-CoV-2 is a key enzyme involved in viral replication and transcription, making it a strong target for drug design. We conducted a systematic study that evaluated the hyperparameters to optimize of these networks specifically for the purpose of analyzing the efficacy of molecular redocking to lower root mean squared deviation for redocking which in turn provides higher confidence binding affinities for analysis of new ligands which could provide pathways for novel drug discoveries.