Improving PLS-RFE based gene selection for microarray data classification.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Microarray data classification faces challenges due to high dimensionality and small sample sizes.
• Existing partial least squares-based gene selection methods are effective but time-consuming.

Purpose of the Study:

• To accelerate the gene selection process for microarray data.
• To improve the efficiency of partial least squares-based recursive feature elimination (PLS-RFE).

Main Methods:

• Integration of PLS-RFE with simulated annealing and square root feature elimination schemes.
• Iterative elimination of multiple features, decreasing with each step.
• Comparison with ReliefF, PLS, and standard PLS-RFE using Naïve Bayes, K-NN, and SVM classifiers on six microarray datasets.

Main Results:

• Proposed methods significantly accelerate the feature selection process.
• Classification accuracy is maintained without degradation.
• More compact gene subsets are obtained for both binary and multi-class problems.
• The simulated annealing scheme shows better time performance and feature subset consistency than the square root scheme.

Conclusions:

• The enhanced PLS-RFE approaches offer a faster and efficient alternative for gene selection in microarray data analysis.
• These methods provide a good balance between computational speed and classification performance.
• The simulated annealing variant demonstrates superior efficiency and consistency.