Identifying differentially expressed genes in unreplicated
multiple-treatment time-course microarray experiments
Rhonda R. DeCook and Dan Nettleton
Department of Statistics
Iowa State University
Microarray technology has become widespread as a means to investigate gene function and metabolic pathways in an organism. A common experiment involves probing, at each of several time points, the gene expression of experimental units subjected to different treatments. Due to the high cost of microarrays, such experiments are often unreplicated. Though an experiment with replication would provide more powerful conclusions, it is still possible to identify differentially expressed genes while controlling the false discovery rate. We present such a method that utilizes polynomial regression models to approximate underlying expression patterns over time for each treatment and gene. Models involving treatment effects, terms polynomial in time, and interactions between treatments and polynomial terms are considered. The "best" model is chosen for each gene using Schwarz's Bayesian Information Criterion. Genes whose "best" model differs significantly from the simplest possible model involving only an overall mean are considered potentially biologically interesting. A two-stage permutation testing approach is used to identify such genes. The expected proportion of false positive results among all positive results is estimated using a method presented by Storey (2001).