A kernel-based clustering method for gene selection with gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High dimensionality and small sample size in gene expression data pose challenges for accurate cancer classification.
• Effective gene selection is crucial for identifying relevant biological markers and improving model performance.

Purpose of the Study:

• To develop and evaluate a new, adaptive clustering-based gene selection method for cancer classification.
• To assess the algorithm's ability to identify potential biomarkers from high-dimensional gene expression data.

Main Methods:

• A novel gene selection approach utilizing kernel-based dissimilarity measures and adaptive distance.
• Iterative optimization of gene weights to enhance clustering objective function performance.
• Testing the algorithm on eight public gene expression datasets using support vector machine and k-nearest neighbor classifiers.

Main Results:

• The proposed gene selection method demonstrated superior accuracy compared to six other feature selection techniques.
• The algorithm proved effective in identifying potential biomarkers across diverse cancer datasets.
• The method showed robustness and simplicity, requiring no dataset-specific parameter tuning.

Conclusions:

• The developed adaptive clustering-based gene selection method is an efficient tool for cancer classification.
• This approach offers a promising strategy for biomarker discovery in gene expression data analysis.
• The algorithm's performance suggests its utility in advancing cancer research and diagnostics.