Interaction-Based Feature Selection for Uncovering Cancer Driver Genes Through Copy Number-Driven Expression Level.

Area of Science:

• Genomics
• Cancer Biology
• Bioinformatics

Background:

• Driver gene identification is crucial for understanding cancer heterogeneity.
• Existing statistical methods often lack biological context.
• Copy number alterations and gene expression levels are key factors in cancer pathogenesis.

Purpose of the Study:

• To develop a novel statistical strategy for cancer driver gene selection.
• To incorporate biological knowledge, specifically the interaction between copy number alterations and gene expression, into driver gene identification.
• To improve the biological relevance and accuracy of identified cancer driver genes.

Main Methods:

• Proposed an interaction-based feature-selection strategy.
• Utilized adaptive L1-type regularization and random lasso procedures.
• Quantified the dependence of copy number alterations on gene expression levels, considering both linear and non-linear effects.

Main Results:

• The proposed method effectively penalizes genes with low feature dependency, leading to smaller or zero coefficients.
• Demonstrated effectiveness in high-dimensional genomic data analysis through Monte Carlo simulations.
• Identified reliable and biologically relevant cancer driver genes using Cancer Genome Atlas (TCGA) data.

Conclusions:

• The novel strategy successfully integrates biological knowledge for more accurate cancer driver gene selection.
• The method provides biologically relevant insights into cancer pathogenesis.
• The approach is effective for analyzing complex, high-dimensional genomic datasets like TCGA.