DPSM-Synergy: A Dual-Path Feature Extraction and Synergy Matrix Enhancement Method for Anti-Cancer Drug Synergy...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Drug synergy prediction is crucial for effective cancer combination therapy.
• Existing computational methods struggle with multimodal drug information and transferable synergy priors.
• Limitations include single feature extraction pathways and ignoring historical synergy and cell line response patterns.

Purpose of the Study:

• To propose a novel prediction model, DPSM-Synergy, to improve anticancer drug synergy prediction accuracy.
• To address limitations of existing models by integrating dual-path feature extraction and synergy matrix enhancement.
• To leverage multimodal drug information and historical synergy patterns for better generalizability.

Main Methods:

• DPSM-Synergy utilizes a dual-path architecture combining a PubChem pretrained model and a graph neural network (GNN).
• It extracts chemical semantic and molecular graph features of drugs.
• Synergy matrices (drug-drug and cell line-drug) are introduced to capture historical synergy patterns and response similarities.

Main Results:

• DPSM-Synergy demonstrated superior performance on DrugCombDB and OncologyScreen datasets.
• It achieved improvements in AUC-ROC by 4.00% and 3.57% over the best baseline on the respective datasets.
• The model validates the effectiveness of the dual-path feature extraction and synergy matrix enhancement strategies.

Conclusions:

• DPSM-Synergy significantly enhances the accuracy of anticancer drug synergy prediction.
• The proposed model effectively integrates multimodal drug information and historical synergy patterns.
• This approach accelerates the discovery of efficient drug combinations for cancer therapy.