kESVR: An Ensemble Model for Drug Response Prediction in Precision Medicine Using Cancer Cell Lines Gene Expression.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Cancer cell lines serve as vital in-vitro tumor models for research.
• Genomic data and large-scale drug screening aid in selecting optimal cancer therapies.
• Accurate drug response prediction is essential for successful precision medicine.

Purpose of the Study:

• To develop and validate a novel computational model for predicting cancer drug response.
• To integrate diverse high-dimensional cancer multi-omics data for improved prediction accuracy.
• To enhance the selection of effective cancer drugs for individual patients.

Main Methods:

• A novel k-means Ensemble Support Vector Regression (kESVR) model was developed.
• kESVR combines supervised and unsupervised learning, utilizing Principal Component Analysis, k-means clustering, and Support Vector Regression.
• The model was trained and validated using gene expression data from the Cancer Cell Line Encyclopedia (CCLE) and Cancer Therapeutics Response Portal (CTRP v2).

Main Results:

• kESVR demonstrated superior accuracy in predicting drug response compared to four standard machine learning models, achieving the lowest mean squared error (MSE).
• In a comparison with 17 other drug response prediction models, kESVR ranked first in 74% of cases and top five in the remaining 26%.
• The model effectively handles missing data and outliers, providing robust predictions.

Conclusions:

• The novel kESVR model offers a significant advancement in predicting cancer drug response from high-dimensional gene expression data.
• kESVR outperforms existing prediction models in both accuracy and speed, while mitigating overfitting.
• This model holds potential for developing robust drug response prediction systems to guide personalized cancer therapy.