Use of relevancy and complementary information for discriminatory gene selection from high-dimensional gene...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technologies generate vast biomolecular data, including gene expression profiles.
• High-dimensional gene expression data presents challenges in identifying key genes (biomarkers) for specific phenotypes like cancer.
• Existing mutual information (MI) methods for gene selection often neglect finite sample correction and dynamic gene discretization, and incorrectly remove potentially important redundant genes.

Purpose of the Study:

• To address limitations in current gene selection methods for high-dimensional gene expression data.
• To propose a novel Mutual Information based Gene Selection (MGS) method that incorporates finite sample correction and dynamic discretization.
• To develop ranking extensions for MGS, namely MGSf (frequency-based) and MGSrf (Random Forest-based), to prioritize selected genes.

Main Methods:

• Developed the Mutual Information based Gene Selection (MGS) method to identify informative genes from gene expression data.
• Incorporated correction for finite sample size in mutual information calculations.
• Introduced dynamic discretization of gene expression data for more relevant gene selection.
• Extended MGS with frequency-based (MGSf) and Random Forest-based (MGSrf) ranking methods.

Main Results:

• The proposed MGS method achieved superior classification rates on diverse gene expression datasets compared to recently reported methods.
• MGS successfully identified key genes associated with disease-relevant pathways.
• The method demonstrated potential in finding responsible genes for unknown disease data.

Conclusions:

• The MGS method offers an effective approach for selecting informative genes from high-dimensional gene expression data.
• MGS and its ranking extensions (MGSf, MGSrf) improve biomarker discovery and classification accuracy.
• The approach holds promise for identifying disease-associated genes and understanding complex biological mechanisms.