Bioinformatics 20(Suppl. 1) © Oxford University Press 2004; all rights reserved.
Predicting gene regulation by sigma factors in Bacillus subtilis from genome-wide data
1 Human Genome Center, Institute of Medical Science, University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan and 2 Graduate School of Biological Science, Nara Institute of Science and Technology, 8916-5 Takayama, Ikoma, Nara 630-0101, Japan
Received on January 15, 2004; accepted on March 1, 2004
Motivation: Sigma factors regulate the expression of genes in Bacillus subtilis at the transcriptional level. We assess the accuracy of a fold-change analysis, Bayesian networks, dynamic models and supervised learning based on coregulation in predicting gene regulation by sigma factors from gene expression data. To improve the prediction accuracy, we combine sequence information with expression data by adding their log-likelihood scores and by using a logistic regression model. We use the resulting score function to discover currently unknown gene regulations by sigma factors.
Results: The coregulation-based supervised learning method gave the most accurate prediction of sigma factors from expression data. We found that the logistic regression model effectively combines expression data with sequence information. In a genome-wide search, highly significant logistic regression scores were found for several genes whose transcriptional regulation is currently unknown. We provide the corresponding RNA polymerase binding sites to enable a straightforward experimental verification of these predictions.
Contact: mdehoon{at}ims.u-tokyo.ac.jp