Irrelevant gene elimination for partial least squares based dimension reduction by using feature probes

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Analyzing high-dimensional gene expression data with few observations is difficult.
• Partial Least Squares based Dimension Reduction (PLSDR) is effective for high-dimensional data.
• Irrelevant features can negatively impact PLSDR accuracy.

Purpose of the Study:

• To develop an improved dimension reduction algorithm for gene expression data.
• To enhance the generalization performance of classifiers using feature selection.
• To address the challenge of analyzing datasets with many genes and few samples.

Main Methods:

• Feature selection was applied to filter gene expression data.
• A novel algorithm, PLSDRg, was developed by integrating PLSDR with gene elimination.
• Gene elimination was guided by t-statistic scores on standardized probes.

Main Results:

• The PLSDRg algorithm demonstrated effectiveness in dimension reduction.
• Experimental results on six microarray datasets confirmed PLSDRg's reliability.
• The method successfully improved the generalization performance of classifiers.

Conclusions:

• PLSDRg offers a robust approach for analyzing high-dimensional gene expression data.
• Integrating feature selection with PLSDR enhances analytical accuracy.
• The algorithm provides a reliable solution for datasets with limited observations and numerous features.