Using bayesian network learning to model yeast transcriptional response to nitrogen oxide
Jingchun Zhu1, Joe DeRisi2
1jzhu@itsa.ucsf.edu, UCSF; 2joe@derisilab.ucsf.edu, UCSF
Motivation:
The Bayesian Network learning technique has been applied in an attempt to infer gene regulatory networks from large-scale gene expression data. We want to apply this technique to model transcriptional response of yeast to nitrogen oxide. Previous work focused on using individual genes as network nodes. In contrast our implementation uses genes clusters as network nodes, which facilitates interpretation of the learning results by reducing the required number of nodes per network. This allows for easier validation of the learned results against the biological hypotheses being explored.
Results:
We used Bayesian Network learning to analyze microarray transcriptional response profiles of yeast to nitrogen oxide. Using gene clusters and experimental conditions, such as the length of time nitrogen oxide was applied, as network nodes, we learned transcriptional response networks from the microarray data. Our learned models, presented as the average of multiple networks, are consistent with the biological hypothesis, using either data of yeast exposed to nitrogen oxide or combined with data of yeast exposed to H202 and menadione. Our results demonstrate that Bayesian Network learning techniques can be used to generate meaningful biological hypotheses, when used with gene expression profiles. In addition, we studied the effect of using different parameters settings. Furthermore, our model revealed a new link between galactose and the expression of a fzf1 dependent gene cluster. We implemented the learning algorithms as a software program.