Abstract: Recent research on large scale microarray analysis has explored the
use of Relevance Networks to find networks of genes that are associated to each
other in gene expression data. In this work, we compare Relevance Networks with
other types of clustering methods to test some of the stated advantages of this
method. The dataset we used consists of artificial time series of Boolean gene
expression values, with the aim of mimicking microarray data, generated from
simple artificial genetic networks. By using this dataset, we could not confirm
that Relevance Networks based on mutual information perform better than
Relevance Networks based on Pearson correlation, partitional clustering or
hierarchical clustering, since the results from all methods were very similar.
However, all three methods successfully revealed the subsets of co-expressed
genes, which is a valuable step in identifying co-regulation.
Keywords: relevance Networks, clustering, regulatory networks, microarray data analysis, simulation