Using functional annotation to improve clusterings of gene expression patterns

The goal of many gene expression experiments is to discover genes that are functionally related by clustering expression levels sampled over some time interval, with the hope that co-regulated genes also are functionally related. However, it is not necessarily always true that co-regulated genes are functionally related, or vice versa, and therefore this paper investigates the value of including gene annotation in the clustering process. Results suggest that clusters formed by a clustering of a combination of expression data and annotation in the form of enzyme classification can give results that have higher correlation with known biological data (functional and metabolic pathway) not included in the clustering process. The results show that the same is true even in a situation with only 10% of the dataset annotated, which is an estimate of the amount of enzymatic annotation available today and a sign that the inclusion of added data helps in the clustering of genes without any explicit annotation. (C) 2002 Elsevier Science Inc. All rights reserved.