Multidrug representation learning based on pretraining model and molecular graph for drug interaction and combination...

Area of Science:

• Computational chemistry and bioinformatics
• Drug discovery and development
• Pharmacology

Background:

• Multidrug therapy is crucial for enhancing treatment efficacy and reducing side effects, but predicting drug interactions remains challenging.
• Existing computational methods for drug interaction identification face limitations such as redundant features, scarce labeled data, and poor generalization.
• There is a need for advanced methods in multidrug representation learning to leverage limited data effectively.

Purpose of the Study:

• To develop a novel pretraining framework, MGP-DR (Molecular Graph Pretraining for Drug Representation), for drug pair representation learning.
• To improve the prediction of drug-drug interactions and drug combinations.
• To address the limitations of existing methods in handling scarce data and redundant features.

Main Methods:

• Integrated unlabeled drug molecular graph and target information using graph and pretraining models.
• Designed a self-supervised learning strategy within the MGP-DR framework.
• Mined contextual information within and between drug molecules for interaction prediction.

Main Results:

• The MGP-DR model demonstrated promising performance across multiple metrics compared to state-of-the-art methods.
• Successfully predicted drug-drug interactions and potential drug combinations.
• The framework effectively utilizes unlabeled data for robust drug representation.

Conclusions:

• MGP-DR offers a reliable approach for learning drug representations and predicting interactions.
• The model provides a valuable candidate set for the combined use of multiple drugs in therapy.
• This framework advances the field of computational drug discovery and personalized medicine.