Encoding Prior Biological Knowledge in Functional Genomics Analysis

Thursday, October 2, 2003 - 11:30am - 11:55am
Keller 3-180
Michael Ochs (Fox Chase Cancer Center)
Cancer is a leading cause of death throughout the world. The fundamental cellular biology underlying the development of cancer is extremely complex, since cancer arises from a myriad of different cellular malfunctions. It is clear, however, that cellular signaling pathways that control cell growth, differentiation, apoptosis, and motility play a critical role in many cancers. New technologies such as microarrays and protein arrays offer the possibility of elucidating key pathways involved in cancer and of monitoring the effect of targeted therapeutics on those pathways. However, because of the limited nature of our knowledge of signaling pathways in humans and high noise levels in the data, difficulties arise during analysis. The inclusion of prior knowledge can enhance probabilistic reasoning in such a case. Analysis of functional genomics data is especially suitable for the inclusion of prior information, since a vast framework of biological knowledge exists.

Bayesian Decomposition is a Markov chain Monte Carlo method that uses Bayesian statistics to encode prior knowledge. The inclusion of biological information both during the analysis and when interpreting patterns identified in the data has greatly increased the power of the algorithm. This is demonstrated with three separate data sets. First, the recovery of a pattern related to the yeast mating pathway is accomplished by use of annotations from the Yeast Proteome Database. Second, tissue identification in Black6 mice is used to isolate tissue specific expression patterns that can be interpreted using gene ontology. Third, links between genes known to be coregulated in yeast is used to demonstrate the effect of such prior knowledge on the analysis.