The 3-phosphoinositide-dependent kinase-1 (PDK1) is a critical target in cancer therapy due to its role in regulating numerous downstream kinases through a protein-protein interaction (PPI) at its PDK1-interacting fragment (PIF) pocket. This allosteric site is an attractive target for developing novel anticancer agents. To facilitate the discovery of modulators for this pocket, we employed a deep learning (DL) approach. The present work developed and evaluated several classification and regression models to predict active compounds. The best-performing model was used to screen the Enamine PPI library of over 29,000 compounds. The top three candidate compounds were docked into the PIF pocket, and their binding modes were compared to the native ligand, PS210. This work led to the identification of small molecules that are predicted to modulate PDK1 activity by binding to the PIF pocket. Subsequent molecular dynamics (MD) simulations confirmed that these compounds bind similarly to the native ligand, maintaining stable interactions within the pocket. The results demonstrate the effectiveness of a deep learning-based approach for predicting inhibitors against the PDK1 PPI. The identified compounds represent promising starting points for developing novel PDK1 modulators. • Deep learning model developed from PDK1 binders. • Virtual screening against Enamine protein-protein interaction inhibitors library. • Three new compounds identified with proven Molecular dynamics and ADMET profile.
Vennila et al. (Sun,) studied this question.