Construction of Gene Regulatory Networks Based on Genetic Algorithm of Greedy Equivalence Search Mechanism

Dynamic Bayesian network (DBN) is an important approach for predicting gene regulatory networks from microarray data. However, three problems greatly reduce the effectiveness of current DBN methods, including long computational time, instable structures, and low accuracy. Here we proposed a method designed to predict gene regulatory networks based on Genetic Algorithm (GA) of Genetic Equivalence Search (GES) Mechanism. According to decomposability of DBN, we divided DBN into initial network and transferring network, then separately encoded and combined them to obtain the chromosome for GA. Two mutation operators are designed based on GES mechanism for GA, made the evolution process of network structure in Markov Equivalence space, rather than in Directed Acyclic Graph (DAG) space. Comparing our result to two other methods and GA with simple mutation operator, our method is proved more efficient. By consulting KEGG, the network structure we predicted obtains biological supports, too.