Statistical Approach for Biologically Relevant Gene Selection from High-Throughput Gene Expression Data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-dimensional gene expression data analysis presents challenges in selecting biologically relevant genes.
• Existing gene selection methods often rely on classification accuracy and can yield redundant or spuriously associated genes.

Purpose of the Study:

• To develop a robust statistical framework for identifying biologically relevant genes from high-dimensional expression data.
• To improve upon existing gene selection methods by integrating filter and wrapper approaches.

Main Methods:

• A novel statistical approach combining support vector machines (SVM) with Maximum Relevance and Minimum Redundancy (mRMR) was developed.
• Gene selection was performed using statistical significance values derived from a nonparametric test statistic and a bootstrap-based subject sampling model.
• The proposed method was rigorously evaluated against nine existing methods using six real crop gene expression datasets.

Main Results:

• The proposed approach demonstrated superior performance in selecting biologically relevant genes compared to existing methods.
• Performance was assessed through subject classification, quantitative trait loci (QTL) analysis, and gene ontology (GO) enrichment.
• The method effectively combined filter and wrapper techniques for enhanced gene selection.

Conclusions:

• The developed statistical approach offers a more effective framework for selecting biologically relevant genes.
• This method enhances the biological interpretability of high-dimensional gene expression data.
• The approach provides a valuable tool for genomics research, improving gene discovery and understanding biological pathways.