SSGraphCPI: A Novel Model for Predicting Compound-Protein Interactions Based on Deep Learning.

Area of Science:

• Computational Chemistry
• Bioinformatics
• Drug Discovery

Background:

• Accurate compound-protein interaction (CPI) prediction is crucial for efficient drug discovery and development.
• Existing deep learning models using 2D molecular graphs often lose vital compound information.
• Graph convolutional neural networks (GCNNs) are commonly used but have limitations in capturing comprehensive molecular features.

Purpose of the Study:

• To develop a novel deep learning framework, SSGraphCPI, for enhanced CPI prediction.
• To integrate sequential and structural features from both 1D SMILES strings and 2D molecular graphs.
• To improve the accuracy and efficiency of identifying potential drug candidates.

Main Methods:

• Proposed a three-channel deep learning framework, SSGraphCPI.
• Utilized recurrent neural networks with an attentional mechanism and graph convolutional neural networks.
• Extracted compound features from both 1D SMILES strings and 2D molecular graphs, incorporating 1D CNN for deeper pattern mining.

Main Results:

• SSGraphCPI achieved significant performance on the GPCR dataset with RMSE = 2.24 and R2 = 0.039.
• The model demonstrated strong predictive capabilities, outperforming classical deep learning models like RNN/GCNN-CNN, GCNNet, and GATNet.
• Experimental results indicate superior performance in capturing comprehensive compound information.

Conclusions:

• SSGraphCPI offers a more effective approach to CPI prediction by leveraging multi-modal compound representations.
• The proposed framework enhances the identification of drug-target interactions, accelerating drug discovery.
• This method provides a valuable tool for virtual screening and drug repurposing efforts.