What is it about?

We describe IPBT, Informative Prior Bayesian Test. IPBT is a model-based method to detect deferentially expressed (DE) genes using microarray data. The key idea in IPBT is the gene-specific informative prior we derived from a large collection of historical microarray data. IPBT outperforms Limma in all real datasets we have tested.

Featured Image

Why is it important?

Bayesian hierarchical model has been the dominating method for detecting deferentially expressed (DE) genes in microarray gene expression data analysis. However, genes are heterogeneous, when borrowing information from genes that are dramatically different from the current one, noise and bias are introduced. This is particularly problematic for genes with naturally low variance. To overcome this problem, in our method named IPBT, we instead borrow strength from the same gene, but from different datasets, as long as it is obtained on the same type of array. There are plenty of public data available for the Affymetrix U133 array, for example. By incorporating historical data under the simple informative prior framework that is natural in the Bayesian paradigm, we provide a simple, yet powerful novel method for detecting DE genes. We showed that this simple idea delivers significant performance enhancement on multiple real data example including the bench mark spike in data provided by Affymetrix.


The availability of rich historical data is an important opportunity provided by the Big Data. We showed in this study, a perfect example that smart utilization of historical data can significantly improve the performance of many key analysis procedures.

Dr Zhaohui S. Qin
Emory University

Read the Original

This page is a summary of: Bayesian inference with historical data-based informative priors improves detection of differentially expressed genes, Bioinformatics, October 2015, Oxford University Press (OUP), DOI: 10.1093/bioinformatics/btv631.
You can read the full text:



The following have contributed to this page