A multi-conformational virtual screening approach based on machine learning targeting PI3Kγ.

Area of Science:

• Biochemistry
• Computational Chemistry
• Drug Discovery

Background:

• Developing selective phosphoinositide 3-kinase gamma (PI3Kγ) inhibitors presents significant challenges due to the protein's unique structural characteristics.
• Targeting PI3Kγ is crucial for various therapeutic applications, necessitating effective drug discovery methods.

Purpose of the Study:

• To develop and validate a virtual screening strategy for identifying novel PI3Kγ inhibitors.
• To enhance the efficiency and accuracy of virtual screening by integrating machine learning and multiple protein conformations.

Main Methods:

• Evaluated the performance of six mainstream molecular docking programs for PI3Kγ systems, identifying CDOCKER and Glide as reliable tools.
• Implemented a virtual screening approach utilizing multiple PI3Kγ protein structures, demonstrating improved enrichment rates compared to single-structure methods.
• Constructed a multi-conformational Naïve Bayesian Classification model integrated with optimal docking programs for inhibitor screening.

Main Results:

• CDOCKER and Glide docking programs exhibited satisfactory reliability and accuracy for PI3Kγ.
• Virtual screening using multiple PI3Kγ protein structures significantly enhanced screening enrichment rates.
• The developed machine learning model demonstrated robust capability in screening PI3Kγ inhibitors.

Conclusions:

• The study successfully developed a validated virtual screening strategy for novel PI3Kγ inhibitors.
• Integrating multiple protein conformations and machine learning improves the discovery of selective PI3Kγ inhibitors.
• This approach provides valuable guidance for future docking-based virtual screening endeavors in drug discovery.