Feature Selection for Breast Cancer Classification by Integrating Somatic Mutation and Gene Expression

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Understanding breast cancer molecular mechanisms is vital for early detection and treatment.
• High-throughput sequencing generates vast data, complicating the identification of driver mutations.
• Identifying a minimal set of critical genes for cancer classification remains challenging.

Purpose of the Study:

• To propose a novel, information-free method for identifying breast cancer-associated mutated genes.
• To develop an integrative approach combining mutation and gene expression data.
• To establish an effective feature selection and evaluation strategy for complex genomic datasets.

Main Methods:

• Processed somatic mutation data into a mutation frequency matrix.
• Filtered a mutated gene set based on mutation frequency thresholds.
• Constructed a co-expression profile by mapping the mutated gene set onto gene expression data.
• Employed staged feature selection including fold change, false discovery rate, mutual information, and gradient boosting decision trees with Bayesian optimization.
• Utilized a weighted metric for evaluation to address sample imbalance.

Main Results:

• Successfully identified a set of mutated genes implicated in breast cancer carcinogenesis.
• The identified genes include known oncogenes and tumor suppressors.
• The integrative network approach outperformed individual gene expression and PMA50 analysis across most metrics.

Conclusions:

• The proposed method effectively integrates multiple data sources to discover breast cancer-associated mutated genes.
• This approach aids in identifying critical genes for cancer classification and understanding molecular mechanisms.
• The findings highlight the potential for improved prediction and treatment strategies through integrated genomic analysis.