A mutual information-based multiple level discretization network inference from time-series gene expression profiles

Abstract

Discovering a genetic regulatory network (GRN) from time series gene expression data plays an essential role in the field of biomedical research. This is because transcriptional regulation is a fundamental molecular mechanism that is involved in almost every aspect of life, from homeostasis to development, from metabolism to behavior, from reaction to stimuli to disease progression. So that many methods have been proposed for inferring GRNs. Among the proposed methods, Boolean networks are widely used. Although the Boolean network models give good results to some extent and are capable of handling data noise, information loss is their main drawback due to the simplicity of data representation, and this leads to the achieved results still being far from optimal. Thus, it is needed to develop an efficient method which can infer large networks with a reliable result in an acceptable run time. In this regard, we propose a new method namely the mutual information based on multiple level discretization network inferences (MIDNI) from time-series gene expression profiles. For each gene in the input network, real-valued gene expression values are discretized into binary or ternary depending on its distribution before feeding to the reference algorithm. We validated MIDNI with four well-known inference methods, DBN, MICRAT, MIBNI, and GENIE3, through extensive simulations on both the artificial discretized and the artificial real-valued gene expression datasets. Our results illustrated that MIDNI significantly outperformed them in terms of both structural and dynamics accuracies. This implies that MIDNI is an efficient tool to reconstruct the gene regulatory networks, particularly, more efficiently for complex and large networks.