Abstract: The clustering algorithm is fundamental for constructing gene regulatory network. From a biological view, a cluster of genes may be regulated and may function similarly. But the clustering algorithm can detect the co-regulation genes only, with the causal relation genes not obtainable. On the contrary, it can get the independent conditional probability between variables based on the Bayesian network model, but its application is limited by the computational complexity. For a target gene, its possible parent gene sets are determined by using clustering technique, and a multi-variable nonlinear regression is utilized to model the predictors. Coefficient of determination (CoD) is employed to compute the probability of selecting a parent gene set from all the possible parent gene sets for the target gene. Based on the conditional local causal relation discovery theorem, gene regulatory network can be constructed. Experimental result show that the feasibility and computational complexity of constructing gene regulatory network with the proposed method are superior to that traditional methods.

Key words: gene regulatory network, gene clustering, local causal relation discovery(LCD)

摘要： 科学的基因聚类方法是构建基因调控网络的前提，但仅以聚类作为构建网络的主要手段只能找到共同调控的基因，不能精确反映基因之间的相互作用过程。贝叶斯网络模型通过基于图的方式求得多变量之间条件独立的概率因果关系，但因其计算复杂性受到应用层面的限制。该文综合考虑几方面因素，在对基因进行聚类基础上，通过对调控关系的预测获得对目标基因的调控基因组，再利用LCD(local causal relation discovery)方法通过限制搜索条件发现基因间的独立关系，进而获得基因调控网络。实验结果表明了该方法的可行性和有效性。

关键词: 基因调控网络, 基因聚类, LCD

CLC Number: 

• TP391.4