Prediction of cancer drug sensitivity using high-dimensional omic features

Area of Science:

• Genomics
• Pharmacogenomics
• Bioinformatics

Background:

• Targeted cancer therapies rely on specific molecular pathways.
• Drugs with shared molecular targets may exhibit common predictive omic features (e.g., mutations, gene expression).
• Analyzing drugs collectively can reveal shared omic predictors and biological insights into drug response.

Purpose of the Study:

• To develop a novel method for high-dimensional bilevel feature selection.
• To identify shared omic features that predict response across groups of drugs targeting the same molecular pathways.
• To address challenges posed by high dimensionality and feature correlation in omic data.

Main Methods:

• Developed a high-dimensional bilevel feature selection method.
• The method accounts for shared predictors among a group of response variables.
• Utilized simulation studies to evaluate performance against existing methods.

Main Results:

• The proposed method demonstrated substantially higher sensitivity and specificity compared to existing approaches.
• Applied the method to two large-scale cancer cell line drug sensitivity datasets.
• Validation within and between studies confirmed the method's efficacy.

Conclusions:

• The new feature selection method effectively identifies shared omic predictors for drugs targeting common pathways.
• This approach offers improved biological insights and robust prediction of drug response.
• The method shows promise for advancing precision oncology and drug development.