Dynamic gene and transcriptional regulatory networks inferring with multi-Laplacian prior from time-course gene microarray data

This paper presents a dynamic gene and transcriptional regulatory network inferring method by using the time-varying autoregressive (TVAR) model. It employs the Li-based regularization terms with spatial sparsity, temporal continuity and proposed multi-Laplacian prior (MLP) for key transcriptional factors (TFs) and their interactions identification. The MLP regularization allows the connections of a gene to be better preserved as a group so that putative TFs can be identified in dynamic gene network. Furthermore, an ADMM-based method is proposed to solve the problem by using the augmented Lagrangian multiplier technique. The simulation using DREAM 4 datasets shows the proposed method performs better than other well-established algorithms for gene network inferring. This enables us to apply the proposed method to a yeast cell cycle microarray datasets containing 215 genes and 17 timepoints more effectively. We are able to identify key genes and gene interactions align well with the natural of yeast cell cycle and related literatures. These suggest that the proposed method can serve as a useful exploratory tool for putative TFs and dynamic gene/TFs networks identification using microarray data.