Fishing Expedition - a Supervised Approach to Extract Patterns from a Compendium of Expression Profiles

Abstract

Reference databases of expression profiles from diverse mutations and chemical treatments of a single assay offer a bird's-eye view of changing expression patterns due to multiple perturbations, Such a compendium of expression profiles has been used to ascertain the roles of previously uncharacterised genes and infer the pathways through which their impact may take place. However, many genes have multiple molecular functions and are involved in different biological processes. The interaction patterns between genes and profiles from two-dimensional hierarchical clustering of such compendium of data could be very complex and often scattered, This makes it difficult to identify and extract all the genes and profiles whose variation in expression levels is closely associated with a particular target function. In this paper, a supervised component analysis approach is proposed in which a small number of profiles and/or genes of known properties are used as "bait" to help "fish out" other profiles and genes from a reference database that are relevant to a particular function of interest. The final cluster analysis and pattern match is then done using a much-reduced data set.