Effectively Identifying Compound-Protein Interaction Using Graph Neural Representation.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate compound-protein interaction (CPI) prediction is vital for drug design and in silico drug discovery.
• Existing machine learning methods often rely on 1D representations, neglecting the inherent graph structure of molecules.

Purpose of the Study:

• To propose GraphCPI, an end-to-end deep learning framework for enhanced CPI prediction.
• To integrate molecular graph topology and protein sequence chemical context for improved modeling.

Main Methods:

• Developed GraphCPI, a novel deep learning framework for CPI prediction.
• Integrated graph neural networks for compound structural information and convolutional neural networks for protein sequence embedding.

Main Results:

• GraphCPI effectively captures complementary structural and contextual information from compounds and proteins.
• Extensive experiments demonstrate the feasibility and competitiveness of GraphCPI against existing methods on multiple CPI datasets.

Conclusions:

• The proposed GraphCPI framework offers a powerful and competitive approach for CPI prediction.
• Integrating molecular graph structures and protein sequence context significantly advances in silico drug discovery capabilities.