UniSyn: a multi-modal framework with knowledge transfer for anti-cancer drug synergy prediction

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in medicine

Background:

• Drug combinations offer improved cancer treatment outcomes, including enhanced efficacy, reduced toxicity, and delayed resistance.
• Predicting synergistic drug combinations remains a challenge due to complex biological interactions.

Purpose of the Study:

• To develop an interpretable deep learning framework (UniSyn) for accurate drug-synergy prediction.
• To leverage knowledge from monotherapy responses to improve combination efficacy prediction.
• To gain mechanistic insights into drug synergy.

Main Methods:

• UniSyn employs a multi-modal deep learning approach with hybrid attention for integrating drug and cell-line features.
• The framework supports multi-task learning for robust generalization across different drug pairs and cell types.
• Performance was validated using multiple synergy scoring metrics.

Main Results:

• UniSyn demonstrated robust generalization to unseen drug pairs and cell types, maintaining consistent performance.
• The model successfully captured context-specific synergy signals when applied to large-scale tumor cell line data.
• UniSyn identified potential therapeutic combinations with translational value.

Conclusions:

• UniSyn provides a powerful and interpretable tool for predicting synergistic drug combinations in cancer therapy.
• The framework's ability to transfer knowledge and provide mechanistic insights facilitates the discovery of novel combination therapies.
• UniSyn has the potential to accelerate the development of more effective and personalized cancer treatments.