Recursive Mahalanobis separability measure for gene subset selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Mahalanobis class separability is crucial for feature selection but computationally intensive for high-dimensional gene expression data.
• Existing methods often lead to overfitting in gene expression analysis due to small sample sizes.

Purpose of the Study:

• To develop a computationally efficient recursive approach for evaluating the Mahalanobis measure.
• To propose a regularized version to mitigate overfitting in gene expression feature selection.

Main Methods:

• A recursive evaluation strategy reducing high-dimensional Mahalanobis calculations to successive 2D evaluations.
• Integration with a forward search procedure for enhanced efficiency.
• Introduction of a regularized recursive Mahalanobis measure with parameter determination guidelines.

Main Results:

• The proposed recursive approach significantly reduces computational overhead.
• The regularized recursive Mahalanobis measure demonstrates superior performance compared to non-regularized methods.
• Experimental results show substantial outperformance against the Recursive Feature Elimination (RFE) benchmark across five gene expression problems.

Conclusions:

• The regularized recursive Mahalanobis measure offers an efficient and effective solution for gene expression feature selection.
• This method addresses computational challenges and overfitting issues inherent in analyzing gene expression data.
• The approach provides a robust alternative to existing feature selection algorithms for biological data analysis.