PLAGCA: Predicting protein-ligand binding affinity with the graph cross-attention mechanism

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Protein-ligand binding affinity prediction is crucial but experimentally costly.
• Existing computational methods often overlook local interaction features, limiting accuracy.
• Accurate prediction accelerates the identification of potential drug candidates.

Purpose of the Study:

• To develop a novel computational method for enhanced protein-ligand binding affinity prediction.
• To integrate global and local features for more accurate binding affinity estimations.
• To address the limitations of existing methods in capturing complex interactions.

Main Methods:

• Proposed PLAGCA (Protein-Ligand binding Affinity prediction using Graph Cross-Attention).
• Utilized sequence encoding and self-attention for global protein/ligand features.
• Employed graph neural networks and cross-attention for local 3D pocket-ligand interactions.
• Integrated global and local features for input into a multi-layer perceptron (MLP).

Main Results:

• PLAGCA demonstrated superior performance compared to state-of-the-art methods.
• The method achieved high accuracy in predicting protein-ligand binding affinity.
• PLAGCA exhibited excellent generalization capability on unseen data.
• Identified critical functional residues contributing to protein-ligand binding.

Conclusions:

• PLAGCA offers a significant advancement in computational protein-ligand binding affinity prediction.
• The integration of local 3D interaction features enhances predictive accuracy.
• The method provides a more efficient and reliable alternative to experimental approaches.