AAPPE: Protein-Ligand Binding Affinity Prediction Leveraging Amino Acid Pair Positional Encoding in Deep Learning.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Protein-ligand binding affinity prediction is crucial for drug discovery.
• Current methods face challenges in efficiently capturing complex interactions.
• Accurate prediction requires integrating structural and chemical information.

Purpose of the Study:

• To develop a novel deep learning framework for predicting protein-ligand binding affinity.
• To integrate spatial relationships between amino acids and ligand features.
• To create a pose-free and computationally efficient prediction model.

Main Methods:

• Introduced amino acid pair positional encoding (AAPPE) framework.
• Integrated spatial relationships of amino acids in protein pockets with ligand molecular fingerprints.
• Encoded pairwise distances of biologically relevant atoms into fixed positional ranges.
• Developed a 3124-dimensional feature set independent of ligand binding poses.

Main Results:

• Achieved robust predictive performance on the CASF-2016 benchmark (MAE = 0.99, RMSE = 1.28, R = 0.82).
• Ablation studies confirmed the significance of biologically informed atom selection.
• Demonstrated computational efficiency and pose-free prediction capabilities.

Conclusions:

• AAPPE offers a generalizable and interpretable approach for structure-based drug design.
• The method provides a practical tool for prioritizing interactions in protein-ligand complexes.
• This framework can accelerate the drug discovery pipeline by improving binding affinity predictions.