Alexander Sturn

Cluster Analysis for Large Scale Gene Expression Studies High throughput gene expression analysis is becoming more and more important in many areas of biomedical research. cDNA microarray technology is one very promising approach for high throughput analysis and provides the opportunity to study gene expression patterns on a genomic scale. Thousands or even tens of thousands of genes can be spotted on a microscope slide and relative expression levels of each gene can be determined by measuring the fluorescence intensity of labeled mRNA hybridized to the arrays. Beyond simple discrimination of differentially expressed genes, functional annotation (guilt-by-association) or diagnostic classification requires the clustering of genes from multiple experiments into groups with similar expression patterns. A platform independent Java package of tools has been developed to simultaneously visualize and analyze a whole set of gene expression experiments. After reading the data from flat files several graphical representations of hybridizations can be generated, showing a matrix of experiments and genes, where multiple experiments and genes can be easily compared with each other. Fluorescence ratios can be normalized in several ways to gain a best possible representation of the data for further statistical analysis. Hierarchical and non hierarchical algorithms have been implemented to identify similar expressed genes and expression patterns, including: 1) hierarchical clustering, 2) k-means, 3) self organizing maps, 4) principal component analysis, and 5) support vector machines. More than 10 different kinds of similarity distance measurements have been facilitated, ranging from simple Pearson correlation to more sophisticated approaches like mutual information. Moreover, it is possible to map gene expression data onto chromosomal sequences. The flexibility, variety of analysis tools and data visualizations makes this software suite a valuable tool in future functional genomic studies.