IHGCN-PLA: An interpretable heterogeneous graph convolutional network for protein-ligand binding affinity prediction...

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Protein-ligand binding affinity prediction is crucial for computer-aided drug discovery.
• Existing graph neural network methods struggle to integrate diverse molecular representations for accurate interaction modeling.

Purpose of the Study:

• To develop an interpretable heterogeneous convolutional network for enhanced protein-ligand binding affinity prediction.
• To introduce an early-stage multimodal interaction fusion mechanism for improved feature learning.

Main Methods:

• Developed an interpretable heterogeneous convolutional network (IHGCN).
• Implemented an early-stage multimodal interaction fusion mechanism integrating structural topology, physicochemical properties, and interaction dynamics.
• Employed heterogeneous graph convolution for cross-modal feature learning.

Main Results:

• Achieved performance comparable to or exceeding full-protein multimodal models using only binding pocket-ligand data.
• Demonstrated state-of-the-art performance on the PDBbind benchmark with reduced data requirements.
• Interpretability analysis confirmed the role of multimodal interaction edges in identifying key binding determinants.

Conclusions:

• The proposed early-stage multimodal fusion strategy effectively models complex protein-ligand interactions.
• The IHGCN framework offers a powerful and interpretable approach for binding affinity prediction.
• The method shows practical value, demonstrated by its application to non-small cell lung cancer.