Combining multiple microarrays in the presence of controlling variables

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Microarray technology allows simultaneous monitoring of thousands of gene expression levels.
• Analyzing data from multiple sources (laboratories, hospitals) requires methods to account for variability.
• Identifying differentially expressed genes is a key objective in microarray experiments.

Purpose of the Study:

• To develop a statistical model for analyzing microarray data from different hospitals.
• To identify differentially expressed genes while accounting for confounding variables like hospital effects.
• To provide a more flexible alternative to meta-analysis for multi-site microarray studies.

Main Methods:

• Extension of the analysis of variance (ANOVA) model to a two-stage approach.
• Stage 1: Adjustment for non-interest effects (e.g., hospital of origin).
• Stage 2: Detection of differentially expressed genes using residuals from Stage 1, employing a permutation test.
• Application to a dataset of 133 microarrays from three hospitals.

Main Results:

• The proposed two-stage ANOVA model effectively accounts for additional variability from confounding factors like different hospitals.
• The permutation test based on residuals successfully identifies differentially expressed genes.
• The model demonstrated flexibility and ease of incorporating individual covariates compared to meta-analysis.

Conclusions:

• The two-stage ANOVA model offers a robust and flexible framework for analyzing multi-site microarray data.
• This approach enhances the accuracy of identifying differentially expressed genes in pooled datasets.
• The developed method provides a practical solution for integrating and analyzing heterogeneous microarray data.