Jae K. Lee, Nitin Jain, HyungJun Cho, and Michael O'Connell

The standard analysis approach to the novel gene discovery in microarray data is emerging as one based on statistical significance and hypothesis testing for each gene's differential expression, with careful attention paid to multiple comparison issues. However, when only a small number of replicated arrays are available, these approaches can often be underpowered and inevitably result both in high false positive and false negative error rates due to their inaccurate within-gene error estimation.

In this lab session, we introduce advanced error information integration methods to overcome such restrictions in small sample microarray data analysis. These are: local pooled error (LPE) method for testing two comparing conditions and empirical Bayes heterogeneous error model (HEM) for simultaneously identifying differentially expressed genes among multiple conditions in a microarray study, especially the latter a natural Bayesian extension of LPE to multiple-condition microarray data. With their resampling-based false discovery rate (FDR) evaluation designed for small sample microarray data, these approaches have been proven to significantly improve statistical discovery power in microarray data analysis with limited replication.

We will guide their basic concepts and the use of these packages with several practical microarray data sets with step-by-step hands-on demonstrations that can be conveniently followed both by computational and biological researchers.