Compound activity prediction using models of binding pockets or ligand properties in 3D

Area of Science:

• Biochemistry
• Structural Biology
• Computational Chemistry

Background:

• Small molecule interactions with proteins are vital for biological processes.
• Advances in protein structure determination and docking enable accurate prediction of these interactions.
• The Pocketome encyclopedia provides data for building ligand activity models.

Purpose of the Study:

• To overview two types of 3D ligand activity models: pocket-based and ligand property-based.
• To evaluate these models for nuclear receptors and G-protein coupled receptors.
• To assess the potential of these models for predicting chemical activity.

Main Methods:

• Utilized data from the Pocketome encyclopedia.
• Developed and applied pocket-based 3D models.
• Developed and applied ligand property-based 3D models.
• Benchmarked model performance using area under the ROC curve (AUC).

Main Results:

• Pocket models achieved acceptable accuracy (AUC > 84%) for half of the targets.
• High accuracy (AUC > 95%) was reached for about one-fifth of targets using pocket models.
• Ligand property field models exceeded 95% performance in 50% of cases.
• Identified strengths and limitations of both modeling approaches.

Conclusions:

• 3D ligand activity models, particularly pocket-based and ligand property-based, can accurately predict small molecule interactions.
• High-performance models are suitable for predicting the activity of novel chemicals.
• Family-wide benchmarking is essential for understanding model capabilities and challenges in drug discovery.