Inferring of regulatory networks from expression data using bayesian networks

Subject of Research. The paper considers the inferring of gene regulatory networks in the form of Bayesian networks from gene expression data. We present this problem as the problem of the marginal probability estimation for each edge appearance in the true Bayesian network under the known gene expression levels. Monte Carlo approach based on the Markov chains is proposed. Method. The proposed method involved the sampling of Bayesian network pairs and a discretization policy, providing a way for the network to be applied to continuous gene expression data according to a posteriori distribution. The Markov chain Monte Carlo approach was used for sampling with implementation via the Metropolis-Hastings algorithm. Then, the desired probabilities were estimated based on the obtained sample. Main Results. The proposed method is tested on simulated data from the DREAM4 Challenges. Comparison with the leaders shows that the developed method quality surpasses the leader among the existing methods, the regularized gradient boosting machines method (RGBM), on some tests and is comparable on the others in view of the results. At the same time, the proposed method is flexible enough and can be adapted to the other types of experimental data. Practical Relevance. The method is applicable in computational biology for research of the gene regulation mechanisms in various processes, including the tumor growth or the immune system operation.