Transcription factor prediction in Bacillus subtilis using stochastic differential equations

Michiel de Hoon1, Sascha Ott2, SunYong Kim, Seiya Imoto, Satoru Miyano, University of Tokyo;, university of Tokyo

Dynamic Bayesian networks have been proposed previously to infer gene regulatory networks from time-course gene expression data. Typically, dynamic Bayesian networks assume an equal time interval between the gene expression level measurements, which may not be the case, and may furthermore prevent us from concatenating several time-course experiments to increase the amount of data from which the network is estimated. Here, we model a gene regulatory network by a system of stochastic differential equations, which can be regarded as a generalization of dynamic Bayesian networks that allows for different time intervals between measurements. The model includes both linear and non-linear interactions between genes. We used synthetic data to validate our model numerically, showing that about fifty measurements or more are needed for a reliable network inference. This underlines the necessity of combining data from several time-course experiments. Applying our model to time-course gene expression measurements of Bacillus subtilis yielded a highly significant prediction for the sigW transcription factor, while the prediction for sigX was marginally significant. The prediction of the transcription factors for the sigY and sigV genes was consistent with existing knowledge, and led to an improved understanding of their regulation.