An adaptation of the LMS method to determine expression variations in profiling data

Area of Science:

• Bioinformatics
• Genomics
• Statistical Modeling

Background:

• Determining differential gene expression thresholds is challenging due to the inverse relationship between variance and signal intensity.
• False positives are a significant concern in expression profiling analysis, impacting the reliability of results.

Purpose of the Study:

• To introduce a novel model, expression variation (EV), for robust gene expression data normalization and analysis.
• To address the issue of variance stabilization and improve the identification of differentially expressed genes.

Main Methods:

• The expression variation (EV) model utilizes the LMS method and cubic spline curves with Box-Cox transformation for data normalization.
• Confidence bands are constructed based on the actual data variance to identify genes with significant variation.
• Dispersion space (DS) and P-values are employed to statistically determine expression variation (EV) for each outlier.

Main Results:

• The EV model demonstrated robust variance stabilization across two Affymetrix datasets.
• Comparison with classical methods showed EV's superior performance in selecting differentially expressed genes.
• EV effectively identified differential expression in both rare and strongly expressed genes.

Conclusions:

• The expression variation (EV) model offers a more robust approach to variance stabilization in gene expression analysis.
• EV enhances the accuracy of differential expression detection, outperforming traditional methods.
• This model provides a reliable tool for analyzing gene expression profiles and identifying biologically relevant changes.