Autoencoder Based Feature Selection Method for Classification of Anticancer Drug Response.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Individual patient responses to anticancer drugs vary significantly due to genetic factors, including mutations and RNA expression levels.
• Genetic features are crucial for developing predictive models of drug response, but the high dimensionality of feature spaces presents a challenge.

Purpose of the Study:

• To address the feature selection challenge in building accurate predictive models for anticancer drug response.
• To develop and evaluate a novel, efficient method for identifying key genetic features that contribute to drug response prediction.

Main Methods:

• An autoencoder network was employed for initial dimensionality reduction and selection of important genetic features.
• The Boruta algorithm was subsequently used to identify a smaller, refined set of features for the predictive model.
• A Random Forest classifier was utilized to predict drug response based on the selected features.

Main Results:

• The proposed method successfully identified a concise subset of genetic features crucial for predicting drug response.
• Validation on the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) datasets demonstrated the method's efficiency.

Conclusions:

• The combined approach of autoencoder networks and the Boruta algorithm offers an effective strategy for feature selection in anticancer drug response prediction.
• This method can enhance the development of personalized medicine by improving the accuracy of predictive models for cancer treatment.