Modelling Gene Networks by a Dynamic Bayesian Network-Based Model with Time Lag Estimation

Abstract

Due to the needs to discover the immense information and understand the underlying mechanism of gene regulations, modelling gene regulatory networks (GRNs) from gene expression data has attracted the interests of numerous researchers. To this end, the dynamic Bayesian network (DBN) has emerged as a popular method in GRNs modelling as it is able to model time-series gene expression data and feedback loops. Nevertheless, the commonly found missing values in gene expression data, the inability to take account of the transcriptional time lag, and the redundant computation time caused by the large search space, frequently inhibits the effectiveness of DBN in modelling GRNs from gene expression data. This paper proposes a DBN-based model (IST-DBN) with missing values imputation, potential regulators selection, and time lag estimation to tackle the aforementioned problems. To evaluate the performance of IST-DBN, we applied the model on the S. cerevisiae cell cycle time-series expression data. The experimental results revealed IST-DBN has decreased computation time and better accuracy in identifying gene-gene relationships when compared with existing DBN-based model and conventional DBN. Furthermore, we expect the resultant networks from IST-DBN to be applied as a general framework for potential gene intervention research.