MDDI-SCL: predicting multi-type drug-drug interactions via supervised contrastive learning.

Area of Science:

• Computational pharmacology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-drug interactions (DDIs) pose significant risks to patient safety and complicate treatment.
• Accurate identification of DDI types is crucial for clinical decision-making and understanding interaction mechanisms.
• Existing computational methods for multi-type DDI prediction require performance improvements.

Purpose of the Study:

• To develop an advanced computational method for predicting multi-type drug-drug interactions (DDIs).
• To improve the accuracy and performance of DDI prediction using a novel supervised contrastive learning approach.
• To provide a tool that aids in avoiding adverse drug events and elucidating DDI mechanisms.

Main Methods:

• Proposed MDDI-SCL, a supervised contrastive learning framework with three-level loss functions.
• Employed a drug feature encoder with self-attention and autoencoder for learning drug-level latent features.
• Utilized multi-scale feature fusion and supervised contrastive loss for learning drug pair-level latent features.
• Integrated a classification module for predicting specific DDI types.

Main Results:

• MDDI-SCL demonstrated superior or comparable performance against state-of-the-art methods across three tasks and two datasets.
• Ablation experiments validated the effectiveness of the supervised contrastive learning strategy.
• Case studies confirmed the practical feasibility and utility of the MDDI-SCL method.

Conclusions:

• MDDI-SCL offers a robust and effective approach for multi-type DDI prediction.
• The method contributes to enhancing patient safety by enabling more accurate identification of potential drug interactions.
• The findings support the advancement of computational pharmacology and personalized medicine through improved DDI prediction.