AI-assisted computational screening and docking simulation prioritize marine natural products for small-molecule PCSK9 inhibition

Abstract

SARS-CoV-2 infection has been associated with long-term cardiovascular complications including myocarditis and heart failure, as well as central nervous system sequelae such as cognitive dysfunction and neuropathy. Proprotein convertase subtilisin/Kexin type 9 (PCSK9), a hepatic protease involved in cholesterol regulation, has shown associations with a spectrum of diseases potentially relevant to these Covid-19 complications, such as atherosclerosis. To identify novel human PCSK9 inhibitors, a custom virtual screening pipeline was developed employing (1) a convolutional neural network-based deep learning model, (2) molecular docking using Schrödinger with Glide scoring function, and (3) molecular dynamics (MD) simulations with Gibbs Free Energy Landscape analysis. The deep learning model was trained on a dataset of known central nervous system, cardiovascular, and anti-inflammatory acting drugs and used to screen the CMNPD database. Docking simulations were performed on shortlisted candidates, followed by MD simulations and free energy landscape analysis to evaluate binding affinities and identify key interaction residues. This multi-step in-silico approach identified promising PCSK9 inhibitor candidates with favorable binding profiles, suggesting that AI-assisted virtual screening can be a powerful tool for discovering novel therapeutic agents.