Balancing false discovery and false negative rates in selection of differentially expressed genes in microarrays

Abstract

Genome-wide mRNA expression profiling using microarrays is widely available today, yet analysis and interpretation of the resulting high dimensional data continue to be a challenge for biomedical scientists. In a typical microarray experiment, the number of biological samples is quite modest compared with the number of genes on a microarray, and a probability of falsely declaring differential expression is unacceptably high without any adjustment for multiple comparisons. However, a stringent multiple comparison procedure can lead to an unacceptably high false negative rate, potentially missing a large fraction of truly differentially expressed genes. In this paper we propose a new "balancing factor score" (BFS) method for identifying a set of differentially expressed genes. The BFS method combines a traditional P value criterion with any other informative factors (referred to as balancing factors) that may help to identify differentially expressed genes. We evaluate the performance of the BFS method when the observed fold change is used as a balancing factor in a simulation study and show that the BFS method can substantially reduce the false negative rate while maintaining a reasonable false discovery rate.