DeepHybridCPI: A hybrid deep learning framework for compound-protein interaction prediction.

Area of Science:

• Bioinformatics
• Computational Drug Discovery
• Artificial Intelligence in Pharmacology

Background:

• Deep learning methods for Compound-Protein Interaction (CPI) prediction are crucial for virtual screening and drug discovery.
• Existing methods using shallow Graph Neural Networks (GNNs) and Convolutional Neural Networks (CNNs) have limitations in capturing global compound structures and long-range protein dependencies.
• Complex deep learning architectures often incur significant computational overhead.

Purpose of the Study:

• To introduce DeepHybridCPI, a hybrid deep learning framework for accurate and efficient CPI prediction.
• To overcome the limitations of existing methods by integrating multiscale compound feature extraction and hybrid protein sequence encoding.
• To provide a computationally efficient and effective tool for accelerating drug discovery processes.

Main Methods:

• A hybrid deep learning framework, DeepHybridCPI, was developed.
• It utilizes a multiscale, densely connected GNN for comprehensive compound feature extraction (local substructures and global topology).
• CNNs combined with Long Short-Term Memory (LSTM) networks are employed for protein sequence analysis (local motifs and long-range dependencies).
• Learned compound and protein representations are fused in a unified latent space for interaction modeling.

Main Results:

• DeepHybridCPI demonstrated superior performance over state-of-the-art methods on benchmark Human and C. elegans datasets.
• The model achieved significant improvements in Area Under the Curve (AUC), Precision, and Recall.
• The framework effectively integrates diverse molecular and sequence information for accurate CPI prediction.

Conclusions:

• Combining multiscale compound representations with hybrid sequence encoders in a unified framework is effective for CPI prediction.
• DeepHybridCPI offers a promising approach for accelerating computational drug discovery.
• The developed framework provides an efficient and accurate solution for predicting compound-protein interactions.