Genetic algorithm-based feature selection with manifold learning for cancer classification using microarray data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray data is crucial for cancer classification but presents challenges due to high dimensionality (large p, small n).
• Effective gene selection is vital for accurate cancer classification from complex microarray datasets.

Purpose of the Study:

• To develop a novel gene selection method for cancer classification using microarray data.
• To address the limitations of existing methods in handling the "large p, small n" characteristic of gene expression data.

Main Methods:

• Proposed Iso-GA, a hybrid method combining Isomap (manifold learning) with Genetic Algorithm (GA).
• Utilized Davies-Bouldin index for evaluating candidate solutions and a probability-based framework to refine gene selection.
• Evaluated performance on eight benchmark cancer microarray datasets.

Main Results:

• Iso-GA demonstrated superior performance compared to other gene selection methods.
• Achieved high classification accuracy with a reduced number of selected genes.
• Effectively captured latent nonlinear structures in gene expression data.

Conclusions:

• The Iso-GA method is effective in selecting a minimal set of critical genes from microarray data.
• Achieves competitive cancer classification performance, highlighting its utility in bioinformatics.