Model selection in a global analysis of a microarray experiment

Area of Science:

• Bioinformatics
• Genomics
• Statistical Modeling

Background:

• Complementary DNA (cDNA) microarray data analysis is crucial for understanding gene expression.
• Joint analysis models, combining information across genes, offer greater power than gene-specific approaches.
• Effective normalization and analysis models are needed to accurately identify differentially expressed genes.

Purpose of the Study:

• To evaluate optimal models for data normalization and analysis in cDNA microarray experiments.
• To identify differentially expressed genes between two muscle types in Avileña Negra Ibérica calves.
• To compare various statistical models considering spatial arrangement, technical variations, and variance heterogeneity.

Main Methods:

• Exploration of three major model groups based on spot arrangement, technical effects (dye, array-block), and gene-specific effects.
• Investigation of three sources of residual variance heterogeneity.
• Model comparison using Bayes factors and cross-validation predictive densities.

Main Results:

• The optimal model incorporated array-block, dye, muscle, and array-dye as systematic effects, with gene-related components as random effects.
• Intensity level was identified as the primary source of heteroscedasticity.
• The best model demonstrated superior goodness of fit and predictive ability, leading to improved detection of differentially expressed genes with a reduced false discovery rate.

Conclusions:

• Accurate modeling of experiment-wide variability is essential for reliable differential gene expression inference.
• The chosen model effectively normalized and analyzed microarray data, even with heterogeneous residual variances.
• Experimental design must carefully consider potential sources of bias to ensure accurate gene expression analysis.