DeepTraSynergy: drug combinations using multimodal deep learning with transformers

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug combinations are crucial for effective cancer treatment, offering enhanced efficacy and selectivity.
• Predicting drug synergy is complex but vital for developing novel therapeutic strategies.

Purpose of the Study:

• To introduce DeepTraSynergy, a deep learning model for predicting drug combination synergy.
• To utilize multimodal data, including drug-target, protein-protein, and cell-target interactions, for improved prediction accuracy.

Main Methods:

• DeepTraSynergy employs transformers for drug feature representation.
• A multitask learning framework predicts drug-target interaction, toxicity, and drug combination synergy.
• Auxiliary tasks (toxicity and drug-target interaction prediction) enhance the primary synergy prediction.

Main Results:

• DeepTraSynergy achieved high accuracy in predicting synergistic drug combinations on DrugCombDB (0.7715) and Oncology-Screen (0.8052) datasets.
• The model outperformed existing classic and state-of-the-art methods.
• Integration of protein-protein interaction networks significantly improved prediction performance.

Conclusions:

• DeepTraSynergy offers a powerful and effective deep learning approach for drug combination synergy prediction.
• The multitask framework and multimodal data integration are key to its superior performance.
• This method holds promise for accelerating the development of effective combination cancer therapies.