Current Innovations in Microarray Analysis A look at two sided clustering and context specific Bayesian clustering Amit Kaushal June 4 2001 Overview To date biologists have used one sided clustering to analyze their data While clustering is informative there is much more in the data that we can learn New techniques such as two sided clustering and context specific Bayesian clustering attempt to make mine more information out of these microarrays Clustering Cluster software released in 1998 by Michael Eisen et al Implements a standard statistical algorithm to determine what genes show the most similar expression pattern over all data How Cluster Works 1 Compute a similarity score between every pair of genes Search for the highest score this represents the most similar pair of genes Combine these genes into one node and compute a similarity score between this node and every other gene node Repeat recursively until there is one node which is the whole tree Benefits of Clustering Cluster has become the industry standard for analyzing a biologist s expression data Why One of the first programs able to process huge quantities of data in microarrays So easy to use even a biologist can use it Acceptable results for a first pass analysis Drawbacks of Clustering But there is a lot more information in the data information that Cluster does not extract Can not determine correlation between subsets of genes and experiments The model is not flexible it can not incorporate any prior knowledge we might have about the genes and their functions ie promoter regions clinical data Innovations in the Works Two sided Clustering Forms subsets of genes and experiments Context specific Bayesian Clustering Flexible intuitive model for gene regulation Two sided Clustering4 More representational of biology in that genes that have common function will act together for the duration of the time that they carry out that function two sided clustering lets you see this subset Algorithm very similar to that of one sided clustering Allows data to be clustered into subsets of genes and experiments Bayesian Statistics 2 3 4 Attempts to look at data and model gene relations based on causal relationships Cluster can only model data based on amount of transcription Bayesian Statistics Relies on a Bayesian Network For examples of basic concepts see citation 2 pages 1 3 BN are very good for describing processes that are locally dependent on each other Advantages over Clustering Much richer than clustering since Bayesian methods given the same data set as a clustering program can discovering causal relationships interactions between genes other than positive correlation and finer intra cluster structure 2 Can incorporate all kinds of information not just mRNA output levels and it makes biological sense Citations Eisen M P Spellman P Brown and D Botstein 1998 Cluster analysis and display of genomewide expression patterns PNAS 95 14863 14868 Friedman N et al Using Bayesian Networks to Analyze Expression Data Friedman N and Y Barash Context Specific Bayesian Clustering for Gene Expression Data Recomb 01 Segal E and B Taskar A Gasch N Friedman and D Koller Rich Probabilistic Models for Gene Expression Bioinformatics vol 1 no 1 1 9
View Full Document
Unlocking...