Multimodal generative neural networks and molecular dynamics based identification of PDK1 PIF-pocket modulators

Abstract

Phosphoinositide-dependent protein kinase-1 (PDK1) is a significant regulator of the AGC family of kinases and plays a vital role in the PI3K-AKT pathway. In the active state, PDK1 is identified as the primary kinase to phosphorylate the T-loop of other AGC kinases by binding with the protein–protein interaction (PPI) region that was identified as the PIF pocket. Therefore, this study attempts to screen compounds generated from deep learning generative neural networks for PDK1 PPI modulators. Both the open and closed forms of PDK1 were utilized in machine learning, and the hits were filtered with parameters like quantitative estimate of drug likeliness (QED) and molecular docking scores. The method was compared with the pharmacophore features based approach. As a result, a novel scaffold was identified as a stable ligand that can bind with many nonbonding interactions in the PIF pocket of PDK1, substantiated by molecular dynamics simulations and binding free energy results.