A new FDR algorithm for differential expression analysis of microarray data

Gregory Grant1, Elisabetta Manduchi2, Christian Stoeckert
1ggrant@pcbi.upenn.edu, CBIL; 2manduchi@pcbi.upenn.edu, CBIL

PaGE (Patterns from Gene Expression) is an algorithm we have developed at CBIL for using statistical methods to assign discrete patterns to gene expression data given replicated experiments from multiple experimental conditions. PaGE confidence measures are based on the False Discovery Rate (FDR), and since version 3.0 has used a permutation algorithm to assess this. Advances in compute power of the PC and advances in algorithms allow an impelmentation for the masses which can remove several troublesome assumptions present in the current implemenations that are in wide use. Our algorithms is based on the Westfall and Young minP stepdown distributions. We are now implementing version 5.0 with the improved statistical alrogithm and a improved interface. PaGE is flexible, handling all types of expression array data (filter, two-channel, Affymetrix). PaGE also performs outlier removal and other preprocessing of the data. This poster will highlight the new interface and the new algorithm, and examples will be given with both real and simulated data, and comparisions are made with other methods. The PaGE web site is: http://www.cbil.upenn.edu/PaGE