The invention of microarray technology has enabled expression levels of thousands of genes to be monitored at once. This modernized approach has created large amount of data to be examined. Recently, gene regulatory network has been an interesting topic and generated impressive research goals in computational biology. Better understanding of the genetic regulatory processes would bring significant implications in the biomedical fields and many other pharmaceutical industries. As a result, various mathematical and computational methods have been used to model gene regulatory network from microarray data. Amongst those methods, the Bayesian network model attracts the most attention and has become the prominent technique since it can capture nonlinear and stochastic relationships between variables. However, structure learning of this model is NP-hard and computationally complex as the number of potential edges increase drastically with the number of genes. In addition, most of the studies only focused on the predicted results while neglecting the fact that microarray data is a fragmented information on the whole biological process. Hence this study proposed a network-based inference model that combined biological knowledge in order to verify the constructed gene regulatory relationships. The gene regulatory network is constructed using Bayesian network based on low-order conditional independence approach. This technique aims to identify from the data the dependencies to construct the network structure, while addressing the structure learning problem. In addition, three main toolkits such as Ensembl, TFSearch and TRANSFAC have been used to determine the false positive edges and verify reliability of regulatory relationships. The experimental results show that by integrating biological knowledge it could enhance the precision results and reduce the number of false positive edges in the trained gene regulatory network.