Evolutionary Algorithms for Finding Interpretable Patterns in Gene Expression Data

Microarray Technology allows us to measure the expression of thousands of genes simultaneously, and under specific conditions. Clustering is the main tool used to analyze gene expression data obtained from microarray experiments. By grouping together genes with the same behavior across samples, resultant clusters suggest new functions for some of the genes. Non-exclusive clustering algorithms are required, as a gene may have more than one biological function. Gene Shaving (Hastie et al. 2000) is a clustering algorithm which looks for coherent clusters with high variance across samples, allowing clusters to overlap. In this paper we present two Evolutionary Algorithm approaches, based on Genetics Algorithms (GA) and Estimation of Distribution Algorithms (EDA), whose aim is to find clusters of similar genes with large between-sample variance. We apply our methods GA-Shaving and EDA-Shaving to S. cerevisiae cell cycle dataset outperforming Gene-Shaving results in terms of quality and size of obtained clusters. Furthermore, we use GO Term Finder (Boyle et al. 2004) to evaluate the biological interpretation of the results. It computes the most statistically significant biological processes associated to every cluster by means of the annotations of the Gene Ontology (Gene Ontology Consortium 2004).