Selecting a classification function for class prediction with gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Class prediction using gene expression is vital for diagnostic and prognostic models.
• Existing classification functions show variable performance across different gene expression datasets.
• Optimal function selection for specific datasets remains an open question.

Purpose of the Study:

• To devise a predictive model for selecting the optimal classification function for class prediction on gene expression data.
• To address the challenge of choosing the most effective classification method for diverse datasets.

Main Methods:

• Simulated gene expression data across various parameters (gene-pair correlations, sample size, variance, differential expression, fold change).
• Built and evaluated ten classifiers using ten classification functions for each simulated dataset.
• Employed linear mixed-effects regression to model classification accuracies based on data characteristics.

Main Results:

• Developed a predictive model linking data characteristics to classification function accuracy.
• The model accurately predicted function performance on real-life datasets, showing positive correlations (0.33-0.82) between predicted and expected accuracies.
• Identified key data features influencing classifier performance.

Conclusions:

• The developed predictive model can guide the selection of optimal classification functions for gene expression data analysis.
• Provides a framework for improving the reliability and accuracy of diagnostic and prognostic models.
• The R package 'SPreFuGED' and source code are available for practical application.