A Two-Stage Method Based on Multiobjective Differential Evolution for Gene Selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray gene expression data are crucial for disease diagnosis and cancer classification.
• Gene selection from high-dimensional data with small sample sizes presents a significant bioinformatics challenge.
• Existing methods often struggle with the complexity of identifying the most informative genes.

Purpose of the Study:

• To propose a novel two-stage gene selection method for microarray data.
• To model the gene selection problem as a multiobjective optimization problem.
• To enhance the accuracy of disease diagnosis and cancer classification through effective gene identification.

Main Methods:

• A two-stage approach combining filter and wrapper feature selection methods.
• Utilizing multiobjective differential evolution (MODE) as the core search strategy in both stages.
• Employing mutual information-based objectives in the filter stage and classification error with feature count in the wrapper stage.

Main Results:

• The proposed method was evaluated on six benchmark gene expression datasets.
• Experimental results demonstrate the effectiveness of the two-stage multiobjective optimization approach.
• The method successfully addresses the gene selection problem in high-dimensional, small-sample datasets.

Conclusions:

• The developed two-stage method offers a novel and effective solution for the gene selection problem.
• Applying multiobjective optimization algorithms significantly improves the identification of informative genes.
• This approach holds promise for advancing disease diagnosis and cancer subtyping using gene expression data.