MGRFE: Multilayer Recursive Feature Elimination Based on an Embedded Genetic Algorithm for Cancer Classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray gene expression data are crucial for cancer classification and bioinformatics research.
• The "large p, small n" problem (high-dimensional data, limited samples) makes gene selection challenging.
• Existing feature selection algorithms often rely on heuristic rules due to computational complexity.

Purpose of the Study:

• To propose a novel multilayer recursive feature elimination (MGRFE) method.
• To select a minimal set of discriminatory genes with maximal information for cancer classification.
• To address the challenges of gene selection in high-dimensional microarray data.

Main Methods:

• Developed a multilayer recursive feature elimination (MGRFE) method.
• Integrated an embedded integer-coded genetic algorithm within the MGRFE framework.
• Evaluated MGRFE on 19 benchmark microarray datasets, including multiclass and imbalanced datasets.

Main Results:

• MGRFE demonstrated superior performance compared to state-of-the-art feature selection algorithms.
• Achieved higher cancer classification accuracy with a smaller number of selected genes.
• Selected genes showed significant biological relevance to cancer phenotypes.

Conclusions:

• MGRFE is a promising feature selection method for high-dimensional datasets, particularly gene expression data.
• The method effectively balances gene selection size and information content.
• The approach offers improved accuracy and biological interpretability for cancer research.