Applying Linear Models to Learn Regulation Programs in a Transcription Regulatory Module Network

The module network method has been widely used to infer transcriptional regulatory network from gene expression data. A common strategy of module network learning algorithms is to apply regression trees to infer the regulation program of a module. In this work we propose to apply linear models to fulfill this task. The novelty of our method is to extract the contrast in which a module's genes are most significantly differentially expressed. Consequently, the process of learning the regulation program for the module becomes one of identifying transcription factors that are also differentially expressed in this contrast. The effectiveness of our algorithm is demonstrated by the experiments in a yeast benchmark dataset.