Document Type

Technical Report

Publication Date






Technical Report Number



Background: One of the most promising but challenging task in the post-genomic era is to reconstruct the transcriptional regulatory networks. The goal is to reveal, for each gene that responds to a certain biological event, which transcription factors affect its transcription, and how several transcription factors coordinate to accomplish specific regulations. Results: Here we propose a supervised machine learning approach to address these questions. We build decision trees to associate the expression level of a gene with the transcription factor binding data of its promoter. From the decision trees, we extract regulatory rules that specify how the binding of a combination of several transcription factors affects the expression of a gene. Such rules are easy to interpret, and represent experimentally testable hypotheses. We use a decision tree ensemble approach to increase modeling accuracy and robustness. We also propose a novel method to integrate rules learned from several time series that measure the same biological processes. We apply our method to publicly available cell cycle expression data and transcription factor binding data for the budding yeast. Cross-validation experiments show that our method is highly accurate and reliable. The method correctly identifies all major known yeast cell cycle transcription factors, and assigns them into appropriate cell cycle phases. It also explicitly reveals synergetic relationships of transcription factors, most of which agree well with existing literatures, while the rest provide testable biological hypotheses. Conclusions: The high accuracy of our method indicates that our method is valid and that the learned regulatory rules can be used as the basic building elements of a transcriptional regulatory network. As more and more gene expression and TF binding data are available, we believe that our method will be useful for reconstructing large scale transcriptional regulatory networks.


Permanent URL: