Identifying Bio-markers for EcoArray

Abstract

EcoArray is a company that makes specialized microarrays for environmental and ecological research. The goal of this project was to rigorously identify contaminants that affect fish, and by extension, humans, using microarray data. This is done by finding genes that are differentially expressed in fish under normal conditions and under treatment by 8 different chemical families PCB-126, Bis-A, Cd, Pb, Hg, Phenantharene, Estradiol and Testosterone. EcoArray uses a proprietary black-box software program called GeneSpring to identify differentially expressed genes that could serve as bio-markers for the 8 different contaminants. The goal of this workshop problem was to develop a statistically robust method of identifying differentially expressed genes given replicate measurements with very different levels of variability.

The top ranking results of our CyberT analysis on the data supplied by EcoArray are markedly different from those obtained earlier by EcoArray using the black box routines implemented in GeneSpring. The output of CyberT on the data are genes that have high differential expression and low variance among the replicates. This suggests that we have implemented a statistically robust method for identifying differentially expressed genes and used it successfully on the data. A further indicator that our method produces more accurate bio-markers is the following: the biomarkers chosen for Cd and Hg by our method have a high degree of overlap and this makes biochemical sense given the similarity between these two toxins - the output produced by GeneSpring did not have this additional signature of consistency.