The Application Of Hidden Markov Model In Building Genetic Regulatory Network

Along with the accomplish of human genes' sequencing, the research for Life Science has gone to post-genome era. The focus of biology's interest has changed from structural genomics to functional genomics. People are more and more interested in the interaction among genes and the construction of genetic regulatory network especially. Building genetic regulatory network can help to explore interaction between genes systematically and understand the natural rule of life phenomenon.For the real genetic regulatory network is a complex and stochastic system, there exist many disadvantages for current deterministic models. For example, the structure of deterministic model is too simple and needs predefined. Besides, deterministic models can't describe genetic regulatory network in an exact way. However, stochastic models are independent of prior knowledge and parametric models. Parameters can be learnt through instance-based learning. This kind of models are more reliable under statistical consideration and more robust by increasing the learning examples. In view of these advantages for stochastic models, probabilistic genetic regulatory network has been paid more attention in recent years and in this paper we mainly discuss its characters and how to build probabilistic genetic regulatory network.Most of the proposed modeling methods for probabilistic genetic regulatory network are based on clustering. In this paper, the widely used k-means clustering algorithm is improved first to enhance clustering efficiency and genetic regulatory network is constructed based on the improved k-means clustering algorithm and redundant clustering. Then, this paper discusses a Hidden Markov Model approach served as a tool to build genetic regulatory network. Different genes clustered into one class are considered as different states firstly and the Hidden Markov Model among these genes can be learnt through relevant algorithms. So the probable regulatory genes for each target gene can be found out through resulting states transition matrix and the probabilistic genetic regulatory network is finally acquired. The experiment based on yeast cell gene expression data proves that the genetic regulatory network reconstructed by the proposed approach can describe interaction among genes and match the experimental data better.