Error-Tolerant Clustering of Gene Microarray Data

Gene microarray technology allows for unprecedented and massive production of biological data across multiple experimental conditions and in time series. Computer analysis of this data can help guide biological bench work toward the assignment of gene function, classification of cells and tissues and the ultimately assist in the diagnosis and treatment of disease. One approach to the analysis of microarray data is the identification of group of genes with common expression patterns or “clusters”. The author implements an error-tolerant clustering algorithm due to Amir Ben-Dor, Ron Shamir and Zohar Yakhini. In their original paper, they defined a stochastic error model for microarray data, and, based on that model, prove that their algorithm recovers the underlying cluster structure of microarray data with high probability. In this paper, their results are replicated on artificial data. In addition, the author tests the stability of clusterings generated by the algorithm and compares the use of discretized and non-discretized similarity graphs. Student: Jay Cahill Degree Candidate 2002, Bachelor of Arts in Computer Science Boston College Email: [email protected] Tel: (617) 308-7218 Advisor: Peter Clote, Ph.D., Doctorat d'Etat Professor of Computer Science Dept of Computer Science and Dept of Biology Boston College Email: [email protected] Tel: (617) 552-1332 Cahill 2 Note: This paper has been modified from its original form. Interest in the original should be directed to Peter Clote, [email protected] or Jay Cahill, [email protected]. Part I: Motivation and Results This paper is organized into three parts. The first part is intended to define the motivation and idea behind the project and present the results. The second part is intended as a user guide and documentation to enable the future maintenance and extension of the CAST software. The third part offers conclusions and ideas for further study. Section