MTFuseSyn: A Multi-Task Fusion Framework for Drug Synergy Prediction Integrating Cell Line Multi-Omics Data

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer incidence and mortality continue to rise, with drug resistance posing a significant challenge in treatment.
• Conventional monotherapy often shows limited efficacy due to tumor cell heterogeneity and multidrug resistance.
• Combination therapy offers enhanced outcomes and delayed resistance, but predicting drug synergism is complex.

Purpose of the Study:

• To develop an advanced computational framework for accurate drug synergy prediction.
• To address the limitations of existing methods in integrating multidimensional data for synergy prediction.
• To enhance cancer treatment strategies through improved understanding of drug combinations.

Main Methods:

• Proposed MTFuseSyn, a multi-task learning framework integrating drug-target and drug-drug interactions.
• Incorporated a graph aggregation module with an adaptive attention mechanism for richer drug features.
• Integrated multimodal cell line data and employed a task fusion attention module for knowledge integration.

Main Results:

• MTFuseSyn significantly outperformed existing methods in both classification and regression tasks on DrugCombDB and Oncology-Screen datasets.
• Demonstrated the effectiveness of multidimensional information fusion for drug synergy prediction.
• Ablation studies and case analyses validated the contribution of proposed modules.

Conclusions:

• MTFuseSyn offers a robust and effective approach for predicting drug synergy in cancer therapy.
• The framework's ability to integrate diverse data sources enhances prediction accuracy.
• This work highlights the potential of advanced computational methods to optimize combination cancer therapies.