AbstractThe regulatory interactions in a cell control cellular response to environmental and genetic perturbations. Gene regulatory network (GRN) inference from high-throughput gene expression data helps to identify unknown regulatory interactions in a cell. One of the main challenges in the GRN inference is to identify complex biological interactions from the limited information contained in the gene expression data. Using prior biological knowledge, in addition to the gene expression data, is a common method to overcome this challenge. However, only a few GRN inference methods can inherently incorporate the prior knowledge and these methods are also not among the best-ranked in benchmarking studies.We propose to incorporate the prior knowledge after the GRN inference so that any inference method can be used. Two algorithms have been developed and tested on the well studied Escherichia coli, yeast, and realistic in silico networks. Their accuracy is higher than the best-ranking method in the latest community-wide benchmarking study. Further, one of the algorithms identifies and removes wrong interactions predicted by the inference methods. With half of the available prior knowledge of interactions, around 970 additional correct edges were obtained and 1300 wrong interactions were removed. Moreover, the limitation that only a few GRN inference methods can incorporate the prior knowledge is overcome. Therefore, a post-inference method of incorporating the prior knowledge improves accuracy, removes wrong edges, and overcomes the limitation of GRN inference methods.
Qualitative reasoning of dynamic gene regulatory interactions from gene expression data