MMFF-DDI: A Multi-Modal Fusion Framework for Drug-Drug Interaction Event Prediction With Contrastive Learning.

Area of Science:

• Pharmacology
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Accurate prediction of drug-drug interaction events (DDIEs) is essential for optimizing combination therapies and enhancing drug safety.
• Existing prediction methods often struggle to integrate local chemical substructures and 3D geometric features due to reliance on limited representations.

Purpose of the Study:

• To develop an advanced multi-modal fusion framework, MMFF-DDI, for improved drug-drug interaction event (DDIE) prediction.
• To overcome the limitations of current methods by jointly capturing diverse molecular features.

Main Methods:

• MMFF-DDI employs a multi-modal approach, extracting drug representations from Morgan fingerprints, canonical SMILES, and 3D molecular graphs.
• Utilizes an attention-augmented autoencoder, MolFormer encoder, and Equivariant Graph Neural Network (EGNN) for feature extraction.
• Incorporates a contrastive multi-modal integration submodule for alignment-based learning, enhancing cross-modal consistency and feature fusion.

Main Results:

• MMFF-DDI significantly outperformed existing methods in predicting DDIEs for existing drugs, showing improvements of 7.87% in Macro-F1 and 7.99% in Macro-precision.
• For novel drug DDIE prediction, MMFF-DDI surpassed competitive methods with 8.06% and 12.79% increases in Macro-F1 and Macro-precision, respectively.
• Visualization and case studies confirmed the framework's practical applicability and superior predictive performance.

Conclusions:

• MMFF-DDI offers a robust and effective solution for predicting drug-drug interaction events by leveraging multi-modal data and contrastive learning.
• The proposed framework enhances the accuracy and reliability of DDIE prediction, contributing to safer and more effective pharmacotherapy.
• The source code is publicly available, facilitating further research and development in this critical area.