Harnessing angular geometry in deep learning for protein-ligand binding affinity prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Protein-ligand binding affinity prediction is crucial for structure-based drug design.
• Current deep learning models often rely on resource-intensive 3D grids or molecular graphs.
• These methods may not fully capture specific directional interactions essential for binding.

Purpose of the Study:

• To introduce novel angular geometric features as descriptors for binding interactions.
• To develop an effective and lightweight model for binding affinity prediction.

Main Methods:

• Extraction of seven types of dihedral angles between protein and ligand atoms to encode orientation and geometry.
• Development of a fully connected ensemble network, the Angle-Aware Predictor (AAP), to integrate these angular features.

Main Results:

• AAP achieved state-of-the-art performance on the CASF-2016 benchmark with R=0.872, RMSE=1.072, MAE=0.817, SD=1.077, and CI=0.845.
• Consistent performance improvements, ranging from 0.3% to 36%, were observed on four additional benchmark datasets.
• The model demonstrates effectiveness, robustness, and computational efficiency.

Conclusions:

• Angular geometric features are effective, lightweight, and robust descriptors for binding affinity prediction.
• Angular geometry represents a promising direction for future structure-based drug discovery.
• The AAP program and data are publicly available for further research and application.