Reconstructing boolean networks from noisy gene expression data

In recent years, a lot of interests have been given to simulate gene regulatory networks (GRNs), especially the architectures of them. Boolean networks (BLNs) are a good choice to obtain the architectures of GRNs when the accessible data sets are limited. Various algorithms have been introduced to reconstruct Boolean networks from gene expression profiles, which are always noisy. However, there are still few dedicated endeavors given to noise problems in learning BLNs. In this paper, we introduce a novel way of sifting noises from gene expression data. The noises cause indefinite states in the learned BLNs, but the correct BLNs could be obtained further with the incompletely specified Karnaugh maps. The experiments on both synthetic and yeast gene expression data show that the method can detect noises and reconstruct the original models in some cases.