Self-Supervised Based Multi-View Graph Presentation Learning for Drug-Drug Interaction Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-drug interactions (DDIs) pose significant risks in polypharmacy, necessitating accurate prediction methods.
• Existing hierarchical graph representation learning for DDIs faces limitations due to scarce experimental data and potential overfitting with supervised methods.
• Supervised models fail to utilize vast unlabeled public molecular datasets, hindering performance.

Purpose of the Study:

• To develop a novel multi-view graph representation learning method, SMG-DDI, for enhanced drug-drug interaction prediction.
• To overcome the data scarcity bottleneck in DDI prediction by leveraging unlabeled molecular datasets.
• To improve the accuracy and generalizability of DDI prediction models.

Main Methods:

• Proposed Self-Supervised Multi-View Graph Representation Learning (SMG-DDI) for DDI prediction.
• Utilized a pre-trained Graph Convolutional Network (GCN) for inter-view molecule graph representation learning (atoms as nodes, bonds as edges).
• Captured intra-view molecular interactions and generated drug embeddings for final DDI prediction.

Main Results:

• SMG-DDI demonstrated superior performance compared to state-of-the-art DDI prediction methods across various dataset scales.
• Achieved prediction accuracies of 0.83, 0.79, and 0.73 on small, medium, and large test datasets, respectively.
• Validated that molecular structure information significantly aids in predicting potential drug-drug interactions.

Conclusions:

• SMG-DDI effectively addresses data limitations in DDI prediction by incorporating self-supervised learning and multi-view graph representations.
• The proposed method offers a promising approach for accurate and reliable prediction of drug-drug interactions.
• Leveraging unlabeled molecular data through advanced graph representation learning enhances DDI prediction capabilities.