Network Completion for Time Varying Genetic Networks

In this paper, we consider the problem of completing and inferring regulatory networks with time varying structure. For this problem, we adopt the methodology of network completion, which is to apply a minimum amount of modifications to given networks so that the resulting network is most consistent with observed data. Network completion can also be applied to network inference by starting with the null network. In order to extend the methodology for completing and inferring time varying network structure, we employ our recent method of network completion, which was obtained by a combination of dynamic programming and least-squares fitting. We extend this method so that edges can be added and deleted at several time points. In order to identify these edges and time points, we develop a novel double dynamic programming method. We perform computational experiments on this method using some artificial data and real expression data.