HOME
Return to Table of Contents
APPLYING META-ANALYSIS TO MICROARRAY: IDENTIFICATION OF A ROBUST GENE SIGNATURE FOR CHRONIC HCV
I Borozan1,5, L Chen1, J Heathcote3, Z Zhang2,5, AM Edwards1,2, MG Katze6, ID McGilvray4,
1Banting and Best Department of Medical Research; 2Department of Medical Genetics and Microbiology; 3Department of Medecine; 4Department of Surgery; 5Terrence Donnelly Center for Cellular and Biomolecular Research, 6University of Toronto and Department of Microbiology, University of Washington
BACKGROUND: Although gene expression profiling is an attractive method for associating alterations in gene expression with clinical outcomes in hepatitis C virus (HCV) infection, the promise of microarray technology has been difficult to realize due in part to variability between different methods (laboratory) and technologies (platform) used. In an effort to identify a high-confidence list of human genes dys-regulated across microarray experiments we have developed a method for combining microarray data across different laboratories, platforms and experimental designs taking into account inter-study variability.
METHODS: We developed a new meta-analysis model based on effect size. This was applied to three independent microarray experiments from two different laboratories that compared normal liver tissue to that obtained from patients with chronic HCV collected from two different laboratories and generated using three different microarray platforms.
RESULTS: Of the 3770 genes that were commonly annotated among the three experiments, data integration produced a set of 206 genes that were increased or decreased in response to HCV infection by a fold change of at least 1.5, with a false discovery rate of 5 %. Of these genes, 79 were found to have higher statistical significance in the combined study than in any of the individual studies. When combined with information from biological pathway databases a number of pathways relevant to HCV were found in the integrated gene set. Two of the enriched pathways from the integrated genes set were not found in any one array dataset, such as the antigen presentation pathway. The genes in this pathway were verified by real-time PCR of liver biopsies from 39 genotype 1 HCV patients.
CONCLUSION: Our analysis demonstrate that data integration not only leads to more reliable results and an increase of statistical power for a sizable number of genes but also identifies genes that could provide concrete insight into clinically relevant mechanisms of host response to hepatitis C viral infection. Our results have formed the basis for a publicly accessible HCV gene expression database.
Funded by NCRTP-HepC