ECCB 2002 Poster sorted by: Author | Number

Next | Previous poster (in order of the view you have selected)
Title: Biological context and the analysis of gene expression data
P10
Apostolakis, Joannis; Güttler, Daniel; Sohler, Florian; Zimmer, Ralf

apostola@informatik.uni-muenchen.de
Ludwigs-Maximillian Universität München

A typical scenario for a DNA chip experiment is the measurement of different samples where the aim is to identify genes that are regulated significantly and bear information on the class of the sample (e.g. healthy or diseased tissue). In a more general setting the question arises of the possibility toextract information on the function of a gene from its expression profile. However, correlation of expression with a particular descriptor, (e.g. the sample class) does not always constitute sufficient indication of biological relations between gene expression and descriptor. At the same time measurements contain significant noise, systematic errors and spurious correlations that together contribute to make extraction of the relevant information rather difficult.

We are investigating simple methods for integrating additional biological knowledge in the analysis of gene expression measurements. In particular we are interested in qualitative gene relation information such as protein interactions, taken from databases or from yeast two hybrid experiments, information derived from text mining, such as coocurrence of gene names in abstracts of scientific publications or existing information on functional similarity of genes. This type of information can in general be described with the help of graphs with genes as nodes and edges denoting relations between genes. The combination of quantitative information (gene expression data) with qualitative information (interaction graphs) is not straightforward. One possible approach is the derivation of a metric from a graph which can then be combined with the quantitative data in some similarity function. The complementary approach is the derivation of correlation graphs from the expression data and the combination of these graphs to obtain new insight.

As a generalization of the prediction of the relevance of genes for different descriptors we are studying the possibility of functional prediction of genes, based on the combination of expression profiles with graphs derived from protein interaction data. We further study the relation between the different types of data, for example the correlation between cooccurrence and expression correlation of the corresponding genes. The discussed approaches and methods are demonstrated and evaluated on the yeast compendium expression data.