Local-Nearest-Neighbors-Based Feature Weighting for Gene Selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Selecting functional genes is crucial for analyzing high-dimensional microarray data.
• Existing feature selection methods, particularly those based on large margin nearest neighbors, face challenges like erroneous neighbor selection and sensitivity to irrelevant genes.

Purpose of the Study:

• To develop a novel local-nearest-neighbors-based feature weighting approach for enhanced gene selection in microarray data analysis.
• To address simultaneous limitations of existing methods, including erroneous nearest neighbors, sensitivity to irrelevant genes, and inappropriate evaluation criteria.

Main Methods:

• Proposed a local-nearest-neighbors-based feature weighting approach.
• Employed a strategy to locally minimize within-class distances and maximize between-class distances.
• Defined and constructed a feature weight vector by minimizing a cost function with a regularization term.

Main Results:

• The proposed approach effectively alleviates problems associated with existing gene selection methods.
• Demonstrated applicability to multi-class problems without requiring modifications.
• Experimental results on UCI and open microarray datasets validated the approach's effectiveness and efficiency.

Conclusions:

• The new local-nearest-neighbors-based feature weighting method offers an effective and efficient solution for gene selection in microarray data.
• The approach successfully handles challenges that previous methods could not address simultaneously, improving the identification of functional genes.