Anticancer drug sensitivity prediction in cell lines from baseline gene expression through recursive feature...

Area of Science:

• Pharmacogenomics
• Computational Biology
• Cancer Research

Background:

• Personalized medicine faces challenges in selecting optimal drugs for individual patients.
• Preclinical drug testing on numerous cell lines is essential due to the impracticality of large clinical trials for novel therapeutics.
• Predictive models are needed to forecast drug efficacy and toxicity efficiently.

Purpose of the Study:

• To develop a predictive model for forecasting anticancer drug response using gene expression data.
• To identify key genomic features that correlate with drug sensitivity.
• To validate the model's performance using independent datasets.

Main Methods:

• Utilized gene expression and drug sensitivity data from the Cancer Cell Line Encyclopedia (CCLE).
• Developed a Support Vector Machine (SVM) based predictor incorporating recursive feature selection.
• Validated model robustness through cross-validation and testing on the independent Cancer Genome Project (CGP) dataset.

Main Results:

• The SVM model demonstrated strong cross-validation performance, achieving over 80% accuracy for 10 drugs and 75% for 19 drugs in CCLE.
• Independent testing on common drugs between CCLE and CGP showed satisfactory prediction for AZD6244, Erlotinib, and PD-0325901.
• The model identified a small set of genes (6-12) crucial for predicting response to these specific drugs.

Conclusions:

• Genomic features can effectively predict drug response, supporting the advancement of personalized medicine.
• The developed model shows potential for predicting drug response for specific agents.
• This predictive approach can complement existing strategies in personalized cancer therapy.