Towards reconstruction of gene networks from expression data by supervised learning

Abstract

Background: Microarray experiments are generating datasets that can help in reconstructing gene networks. One of the most important problems in network reconstruction is finding, for each gene in the network, which genes can affect it and how. We use a supervised learning approach to address this question by building decision-tree-related classifiers, which predict gene expression from the expression data of other genes.

Results: We present algorithms that work for continuous expression levels and do not require a priori discretization. We apply our method to publicly available data for the budding yeast cell cycle. The obtained classifiers can be presented as simple rules defining gene interrelations. In most cases the extracted rules confirm the existing knowledge about cell-cycle gene expression, while hitherto unknown relationships can be treated as new hypotheses.

Conclusions: All the relations between the considered genes are consistent with the facts reported in the literature. This indicates that the approach presented here is valid and that the resulting rules can be used as elements for building and explaining gene networks.

Figure 1

The decision tree for gene CLN2 of S. cerevisiae. Here CLN2 is the predicted gene; SWI5, CLN1 and CDC28 are the explaining genes. Expression thresholds of the respective explaining genes mark all the arcs.

Figure 2

The network of gene interactions constructed using the decision rules for the cdc15 dataset (see Table 2). The network is a graphical representation of the information comprised in the extracted decision rules. Every node in this graph represents a gene and every arc indicates the relation between the genes defined by the corresponding decision rule. Note the existence of two separate modules in the constructed network.