Hierarchical Bayesian meta-analysis models for cross-platform microarray studies

Abstract

The development of new technologies to measure gene expression has been calling for statistical methods to integrate findings across multiple-platform studies. A common goal of microarray analysis is to identify genes with differential expression between two conditions, such as treatment versus control. Here, we introduce a hierarchical Bayesian meta-analysis model to pool gene expression studies from different microarray platforms: spotted DNA arrays and short oligonucleotide arrays. The studies have different array design layouts, each with multiple sources of data replication, including repeated experiments, slides and probes. Our model produces the gene-specific posterior probability of differential expression, which is the basis for inference. In simulations combining two and five independent studies, our meta-analysis model outperformed separate analyses for three commonly used comparison measures; it also showed improved receiver operating characteristic curves. When combining spotted DNA and CombiMatrix short oligonucleotide array studies of Geobacter sulfurreducens, our meta-analysis model discovered more genes for fixed thresholds of posterior probability of differential expression and Bayesian false discovery than individual study analyses. We also examine an alternative model and compare models using the deviance information criterion.

Keywords:

• Bayesian statistics
• meta-analysis
• microarray data
• multiple platform
• Markov chain Monte Carlo
• deviance information criterion

Acknowledgements

This research was supported by the Office of Science (BER), US Department of Energy, Grant No. DE-FC02-02ER63446. We thank three anonymous referees for their thoughtful comments which enhanced the manuscript.