Anticancer drug synergy prediction based on CatBoost.

Area of Science:

• Oncology
• Computational Biology
• Pharmacology

Background:

• Multi-targeted combination drugs are ideal for cancer treatment.
• Exploring drug combinations is challenging due to vast combinatorial space.
• Machine learning offers an effective approach to navigate this space.

Purpose of the Study:

• To develop a machine learning model for predicting anticancer drug synergy.
• To utilize the CatBoost algorithm for enhanced prediction accuracy.
• To identify key drug and cell line features influencing drug synergy.

Main Methods:

• A CatBoost machine learning model was developed to predict synergy scores.
• The model was trained and tested on the NCI-ALMANAC dataset.
• Drug features included Morgan fingerprints, target information, and monotherapy data; cell lines were characterized by gene expression profiles.

Main Results:

• The CatBoost model achieved high performance with ROC AUC of 0.9217 and PR AUC of 0.4651.
• The model significantly outperformed three other advanced methods.
• SHapley Additive exPlanations (SHAP) revealed drug features and specific genes (PTK2, CCND1, GNA11) as crucial for synergy prediction.

Conclusions:

• The proposed machine learning model demonstrates strong predictive capabilities for anticancer drug combinations.
• Drug features were found to be more influential than cell line features in predicting synergy.
• This method serves as a viable alternative for predicting synergistic anticancer drug combinations.