An Improved Anticancer Drug-Response Prediction Based on an Ensemble Method Integrating Matrix Completion and Ridge...

Area of Science:

• Computational biology
• Pharmacogenomics
• Machine learning in oncology

Background:

• Predicting anticancer drug response is crucial for personalized cancer therapy.
• Existing models often lack comprehensive integration of diverse biological data.
• Accurate prediction aids in optimizing treatment strategies and drug development.

Purpose of the Study:

• To develop and validate an ensemble learning method for predicting anticancer drug response.
• To compare the proposed model against a state-of-the-art dual-layer network model.
• To enhance the biological interpretability of drug-response prediction.

Main Methods:

• Ensemble learning integrating low-rank matrix completion and ridge regression.
• Application to Cancer Cell Line Encyclopedia (CCLE) and Genomics of Drug Sensitivity in Cancer (GDSC) datasets.
• 10-fold cross-validation for head-to-head comparison with existing models.

Main Results:

• The proposed model achieved higher prediction accuracy (Pearson correlation) for most drugs across CCLE and GDSC datasets.
• Demonstrated superior performance in predicting drug responses within specific pathways like PI3K and ERK.
• Case studies validated the model's effectiveness and highlighted its biological interpretability.

Conclusions:

• The ensemble learning model provides a robust and accurate approach for predicting anticancer drug response.
• The model's biological interpretability, identifying key genes and functions, aids in understanding drug mechanisms.
• This method holds promise for advancing precision oncology and accelerating drug discovery efforts.