A new regularized least squares support vector regression for gene selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene selection from microarray data is challenging due to high dimensionality and small sample sizes.
• Existing methods often assume equal subject contribution, potentially overlooking disease-related dependencies.
• This limitation can restrict the identification of truly influential genes.

Purpose of the Study:

• To develop a novel gene selection approach that accounts for varying subject contributions.
• To improve the accuracy and efficiency of identifying influential genes in complex diseases.
• To extend the method for both binary and multiclass classification problems.

Main Methods:

• A new gene selection strategy based on kernel similarities and calculated kernel weights.
• Utilizing regularized least squares support vector regression (RLS-SVR) to determine subject weights.
• Ranking genes based on the cumulative sum of weighted expression levels.

Main Results:

• The proposed method successfully identified influential genes across multiple cancer datasets (leukemia, colon, breast, lung).
• It demonstrated effectiveness in both binary and multiclass classification scenarios.
• The approach yielded a smaller set of genes with higher classification accuracy compared to other methods.

Conclusions:

• The kernel weight-based approach is computationally efficient and easy to implement.
• It provides a more accurate gene set by considering individual subject contributions.
• This method offers a significant improvement over traditional gene selection techniques.