Improving therapeutic synergy score predictions with adverse effects using multi-task heterogeneous network learning.

Area of Science:

• Pharmacology
• Computational Biology
• Bioinformatics

Background:

• Drug combinations can yield therapeutic effects (TEs) and adverse effects (AEs).
• Existing computational methods often predict TEs and AEs separately, neglecting shared mechanistic insights.
• Understanding drug-drug interactions (DDIs) is crucial for predicting combination outcomes.

Purpose of the Study:

• To develop a computational method that leverages shared mechanistic commonalities between TEs and AEs for improved TE prediction.
• To test the hypothesis that joint learning of TEs and AEs enhances the prediction accuracy of therapeutic effects.
• To provide a deeper understanding of the mechanisms of action (MoAs) in drug combinations.

Main Methods:

• Formulated TE prediction as a multi-task heterogeneous network learning problem.
• Proposed Muthene (multi-task heterogeneous network embedding) to perform TE and AE learning tasks simultaneously.
• Evaluated Muthene on a drug-drug interaction dataset containing both TE and AE indications.

Main Results:

• Muthene achieved more accurate TE predictions compared to standard single-task learning methods.
• Including AE prediction as an auxiliary task significantly improved TE prediction accuracy.
• The method demonstrated its utility in understanding MoAs for specific drug pairs, like Vincristine-Dasatinib.

Conclusions:

• Jointly learning therapeutic and adverse effects improves the prediction of drug combination outcomes.
• Muthene offers a novel approach to TE prediction by integrating AE data.
• The method enhances understanding of drug combination mechanisms of action (MoAs).