SDDSynergy: Learning Important Molecular Substructures for Explainable Anticancer Drug Synergy Prediction.

Area of Science:

• Computational biology
• Pharmacology
• Drug discovery

Background:

• Drug combination therapies are vital in cancer treatment, offering reduced toxicity and improved efficacy.
• Existing computational methods for predicting drug synergy often overlook the crucial role of specific molecular substructures.
• Identifying key functional groups within drugs is essential for understanding and predicting synergistic effects.

Purpose of the Study:

• To develop a substructure-aware computational method, SDDSynergy, for predicting anticancer drug synergy.
• To adaptively identify critical functional groups contributing to synergistic drug interactions.
• To improve the accuracy and efficiency of discovering novel synergistic drug combinations.

Main Methods:

• SDDSynergy predicts drug synergy by evaluating the effects of individual substructures on cancer cell lines.
• A novel drug-cell line attention mechanism highlights the impact of important substructures.
• A substructure pair attention mechanism captures interactions between substructure pairs within drug combinations, utilizing multilayer substructure information passing networks on molecular graphs.

Main Results:

• SDDSynergy demonstrated superior performance compared to state-of-the-art methods across three real-world datasets.
• The method effectively identifies critical substructures and their contributions to drug synergy.
• Literature surveys confirmed that many novel drug combinations predicted by SDDSynergy are supported by existing research and clinical trials.

Conclusions:

• SDDSynergy represents a significant advancement in computational drug synergy prediction by incorporating substructure-level analysis.
• The method accelerates the identification of effective and potentially safer anticancer drug combinations.
• This approach holds promise for optimizing combination therapy strategies in oncology.