Picture-word order compound protein interaction: Predicting compound-protein interaction using structural images of...

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Identifying compound-protein interactions (CPI) is crucial but challenging in drug discovery.
• Current deep learning methods often represent compounds and proteins as sequences or graphs.
• Computer vision offers potential for richer feature extraction from compound structures.

Purpose of the Study:

• To develop a novel deep learning model for compound-protein interaction prediction.
• To leverage structural images of compounds for enhanced feature representation.
• To improve the accuracy and efficiency of drug discovery pipelines.

Main Methods:

• Proposed the picture-word order compound protein interaction (PWO-CPI) model.
• Utilized convolutional neural networks (CNNs) to learn representations from compound structural images and protein sequences.
• Employed generative adversarial networks (GANs) for feature visualization.

Main Results:

• PWO-CPI significantly outperformed existing state-of-the-art compound-protein interaction prediction models.
• Experiments validated the efficacy of molecular structure images as effective molecular features.
• Visualization confirmed PWO-CPI's ability to implicitly learn compound structural features.

Conclusions:

• Compound structural images provide valuable features for predicting compound-protein interactions.
• The PWO-CPI model offers a promising advancement in computational drug discovery.
• This approach has potential applications in drug-drug interaction studies and beyond.