Novel gene sets improve set-level classification of prokaryotic gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Set-level classification of gene expression data is a recent focus.
• Dimensionality reduction in gene sets aims to reduce overfitting and improve classifier accuracy.
• Previous studies have not consistently confirmed improved accuracy, potentially due to unsuitable gene set definitions.

Purpose of the Study:

• To investigate an alternative approach for defining gene sets based on regulatory interactions.
• To hypothesize that gene sets derived from regulatory interactions will exhibit higher gene expression correlation.
• To test if these more correlated gene sets improve classifier accuracy in set-level analysis.

Main Methods:

• Defined two families of gene sets using regulatory interaction information.
• Evaluated gene sets on phenotype-classification tasks using public prokaryotic gene expression data.
• Selected the best-performing subtype from each family and evaluated them against state-of-the-art and conventional gene-level approaches on independent test datasets.

Main Results:

• The novel gene sets demonstrated higher correlation among genes compared to conventional gene sets.
• Classifiers built using the novel gene sets achieved significantly higher accuracy.
• The improved performance was consistent across evaluations on independent datasets.

Conclusions:

• Gene sets defined by regulatory interactions enhance set-level classification of gene expression data.
• This approach offers a more effective strategy for analyzing gene expression patterns.
• Reproducible experimental materials are provided for the novel gene set methodology.