MDL-CPI: Multi-view deep learning model for compound-protein interaction prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Compound-Protein Interactions (CPIs) are crucial for drug discovery.
• Current computational methods for CPI prediction have limitations, often neglecting interaction information.
• There is a need for improved predictive models to accelerate drug development.

Purpose of the Study:

• To propose a novel multi-view deep learning method, MDL-CPI, for enhanced CPI prediction.
• To address the limitations of existing methods by incorporating interactive information.
• To improve the accuracy and efficiency of computational drug discovery pipelines.

Main Methods:

• Utilized BERT (Bidirectional Encoder Representations from Transformers) and CNN (Convolutional Neural Network) for sequential protein feature extraction.
• Employed GNN (Graph Neural Networks) for structural compound feature extraction.
• Integrated features using an AE2 (Autoencoder in Autoencoder Networks) to learn compound-protein interactive information.

Main Results:

• MDL-CPI demonstrated superior performance compared to existing CPI prediction methods on benchmark datasets.
• The model's effectiveness was validated, showing strong predictive capabilities.
• The study confirmed the critical role of learned interactive information in improving prediction accuracy.

Conclusions:

• MDL-CPI offers a significant advancement in predicting compound-protein interactions.
• Integrating multi-view features and interaction information is key to enhancing predictive models.
• The developed model and released dataset can accelerate drug discovery and development efforts.