Edge-enhanced interaction graph network for protein-ligand binding affinity prediction

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Protein-ligand interactions are fundamental to drug discovery and development.
• Accurate prediction of binding affinity is critical for efficient drug candidate screening.
• Graph neural networks (GNNs) excel at capturing complex spatial and structural information in molecular interactions.

Purpose of the Study:

• To introduce EIGN, a novel GNN-based model for predicting protein-ligand binding affinity.
• To evaluate EIGN's predictive accuracy and generalization capabilities on benchmark datasets.
• To validate the model's effectiveness through comprehensive experimental analyses.

Main Methods:

• EIGN utilizes a three-component architecture: a normalized adaptive encoder, a molecular information propagation module, and an output module.
• The model leverages graph representations to learn intricate patterns in protein-ligand complexes.
• Performance was assessed using standard metrics like root mean squared error (RMSE) and Pearson correlation coefficient (PCC).

Main Results:

• EIGN achieved an RMSE of 1.126 and a PCC of 0.861 on the CASF-2016 dataset.
• The model outperformed state-of-the-art methods on CASF-2013, CASF-2016, and CSAR-NRC datasets.
• Ablation studies and feature importance analyses confirmed EIGN's robust performance and applicability.

Conclusions:

• EIGN demonstrates high accuracy and strong generalization ability in predicting protein-ligand binding affinity.
• The developed GNN model offers a promising tool for accelerating drug discovery pipelines.
• Further validation experiments underscore the model's potential for real-world applications in computational drug design.