Fusing Micro- and Macro-Scale Information to Predict Anticancer Synergistic Drug Combinations

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Drug combination therapy is crucial for cancer treatment.
• Computational methods accelerate the exploration of drug combinations.
• Current deep learning models struggle with generalization in predicting drug synergy.

Purpose of the Study:

• To develop a novel method, MMFSynergy, for accurate prediction of synergistic anticancer drug combinations.
• To address limitations in characterizing drugs and cell lines at micro- and macro-scales.
• To effectively integrate cross-scale information for improved synergy prediction.

Main Methods:

• MMFSynergy utilizes micro and macro encoders for feature extraction from drugs and proteins.
• Cross-scale information fusion is achieved through a self-supervised task.
• A Transformer Encoder model predicts synergy scores using fused drug and cell line protein features.

Main Results:

• MMFSynergy demonstrated superior generalization ability compared to eight advanced methods across two datasets.
• The method effectively captures and integrates micro- and macro-scale features.
• Ablation and case studies confirmed the effectiveness of MMFSynergy.

Conclusions:

• MMFSynergy offers a significant advancement in predicting synergistic drug combinations for cancer therapy.
• Integrating cross-scale information is key to improving model performance.
• The proposed method provides a more effective computational tool for drug discovery.