A multi-view feature representation for predicting drugs combination synergy based on ensemble and multi-task...

Area of Science:

• Computational Biology and Bioinformatics
• Pharmacology and Drug Discovery
• Machine Learning in Medicine

Background:

• Identifying synergistic drug combinations is crucial for effective cancer therapy.
• Current methods often struggle with the complexity of drug-cell line interactions.
• Predictive modeling offers a promising avenue for accelerating the discovery of synergistic drug pairs.

Purpose of the Study:

• To develop and validate a novel multi-view ensemble predictor model (MVME) for synergistic drug combinations.
• To predict both the quantitative synergy score and the qualitative synergy class label for drug pairs.
• To leverage diverse molecular and cellular data views for improved prediction accuracy.

Main Methods:

• Represented drugs using four views: SMILES, molecular graphs, fingerprints, and drug-target interactions.
• Captured cell line features through four views: gene expression, copy number, mutation, and proteomics.
• Employed a multi-task attention deep learning model integrating sixteen paired drug-cell line views, followed by an ensemble model for final prediction.

Main Results:

• The MVME model achieved a Pearson score of 0.76 for synergy score prediction (RMSE: 14.30).
• For synergy class prediction, the model demonstrated high performance with 0.90 accuracy, 0.96 precision, and 0.96 ROC-AUC.
• Evaluated on the O'Neil dataset comprising 22,737 drug combinations across 39 cancer cell lines.

Conclusions:

• The proposed multi-view ensemble approach effectively predicts synergistic drug combinations.
• Integrating multiple feature representations enhances the model's ability to capture complex drug-response relationships.
• This methodology holds significant potential for advancing precision oncology and personalized medicine.