Regression approaches for microarray data analysis.

Area of Science:

• Bioinformatics
• Genomics
• Statistical Genetics

Background:

• Microarray studies present challenges like high gene numbers versus limited samples and correlated gene expression due to biological pathways.
• These challenges are amplified in regression analyses aiming to link multiple gene expressions to external phenotypes.

Purpose of the Study:

• To evaluate existing and novel procedures for two-sample gene expression comparisons using microarrays.
• To assess the effectiveness of gene harvesting and regularized regression methods in handling high-dimensional microarray data.

Main Methods:

• Critique of existing gene expression analysis methods.
• Detailed evaluation of the gene harvesting technique.
• Application and comparison of regularized regression procedures: lasso, least angle regression, and support vector machines.
• Analysis of model selection and solution multiplicity.

Main Results:

• Gene harvesting, without constraints, can produce artifactual results.
• Regularized regression methods, when appropriately constrained, offer more reliable solutions.
• The study highlights the importance of addressing model selection and solution multiplicity in microarray data analysis.

Conclusions:

• Regularized regression procedures are essential for robust analysis of high-dimensional, correlated gene expression data from microarrays.
• Careful consideration of constraints and regularization is necessary to avoid artifacts in gene expression analysis.
• The findings are validated using a mouse model of cardiomyopathy.