AI-assisted molecular docking and molecular dynamics simulations for predicting off-target effects of AKT1 ATP-competitive inhibitors

AKT1 is a member of the AGC kinase family and represents a key therapeutic target in cancer. Although ATP-competitive compounds can act as potent inhibitors of AKT1, they may also exhibit off-target interactions with other kinases in the PI3K/AKT/mTOR pathway, a critical issue that requires systematic investigation. To address this challenge, we propose a computational framework that integrates a deep learning (DL) model with physics-based modeling approaches. Specifically, the developed DL model, CrossAtt-DTI, was first employed to perform binary classification and identify potential AKT1 binders. Subsequently, multiple molecular docking tools were used to predict binding conformations and identify poses that capture key binding residues. The stability of these docked conformations was then evaluated through all-atom molecular dynamics simulations in explicit solvent, followed by MM/PBSA calculations. The proposed framework was initially validated and subsequently applied to investigate the off-target interactions of ATP-competitive inhibitors with other kinases. The results indicate that, in addition to off-target interactions with members of the AGC kinase family predicted for most ATP-competitive inhibitors, Ipatasertib and NTQ1062 may exhibit strong interactions with PI3Kα, a member of the PI3K kinase family, while NTQ1062 may also interact with mTOR, a member of the PI3K-related kinase family. However, further in vitro and in vivo studies are required to validate these potential interactions. Overall, this work establishes a hybrid deep learning and physics-based computational framework for predicting the off-target effects of ATP-competitive AKT1 inhibitors and provides mechanistic insights into kinase cross-reactivity within the PI3K/AKT/mTOR signaling pathway.