A deep learning-based theoretical protocol to identify potentially isoform-selective PI3Kα inhibitors

Area of Science:

• Oncology
• Computational Chemistry
• Pharmacology

Background:

• Phosphoinositide 3-kinase alpha (PI3Kα) is frequently dysregulated in cancer, driving oncogenesis.
• Developing isoform-selective PI3Kα inhibitors is challenging due to structural similarities among PI3Ks.
• Existing PI3Kα inhibitors have associated side effects, necessitating novel therapeutic strategies.

Purpose of the Study:

• To design a hybrid computational protocol integrating machine learning (ML) and in silico drug design for identifying novel PI3Kα inhibitors.
• To overcome the challenge of isoform selectivity in PI3Kα inhibitor development.
• To identify potential drug candidates with improved efficacy and safety profiles.

Main Methods:

• Developed and trained a deep learning classification model on physicochemical descriptors of known PI3Kα inhibitors.
• Screened a small molecule database using the ML model to predict potential inhibitors.
• Applied multiphase molecular docking to assess binding affinities and modes of predicted compounds.
• Conducted drug-likeness, pharmacokinetic (ADME-T), and mechanistic dynamic studies.

Main Results:

• Predicted 662 potential PI3Kα inhibitors using the ML classification model.
• Identified 12 compounds with high binding affinities and favorable interactions with PI3Kα active site residues via molecular docking.
• Selected six compounds (1, 2, 3, 6, 7, 11) with suitable ADME-T profiles and bioavailability.
• Mechanistic studies indicated compounds 1, 2, and 11 bind to the ATP-binding site with higher affinity than alpelisib.

Conclusions:

• The developed ML model and computational strategy are reliable for identifying novel, isoform-selective PI3Kα inhibitors.
• Compounds 1, 2, and 11 demonstrate significant potential for PI3Kα targeted cancer therapy.
• This approach offers a promising avenue for discovering new anti-cancer drug candidates with improved therapeutic indices.