文档介绍：
基于因子图的基因调控网络建模与分析#
谢雪英,李鑫*
(东南大学学习科学研究中心,南京,210096)
5
10
摘要:基因调控网络的研究是从基因间相互作用关系的角度揭示生命体中复杂的生命现象,
有助于对生命过程进行详细的解释,包括细胞活动、生命活动、遗传疾病与治疗等,其研究
成果具有重要的理论意义和应用价值。本文采用一种基于因子图模型的基因调控网络建模方
法,利用消息传递机制进行互作概率推断,研究不同网络参数选取方法对算法的影响,并采
用模拟数据和酵母表达数据对模型性能进行了测试。结果表明,该模型对模拟数据具有非常
好的预测效果,从而很好地验证了该模型的有效性;在真实酵母表达数据中预测中的作用关
系与数据库中已记录的调控关系具有很好的一致性。基于该模型预测的候选调控关系可以为
将来的后续实验研究提供指导。
关键词:基因调控网络;概率图模型;因子图;消息传递算法
中图分类号:,Q61
15
The analysis of gene work based on factor
graph
XIE Xueying1, LI Xin2
(1. The Research Center of Learning Center, Southeast University, Nanjing, 210096;
20
25
30
35
40
2. The Research Center of Learning Science, Southeast University, Nanjing, 210096)
Abstract: With the availability of various types of high-throughput experimental data, in silico
reconstruction of gene works es one of the most challenging issues in functional
genomics. Based on the factor graph and message passing algorithm, we proposed a generalized
pipeline for the construction and analysis of gene works from the gene expression data. In
order to assure the model able to handle the manifold diversity of the data, we put forward two methods
to automatically determine the value of model parameters, and attempt to avoid the uracies by
subjective choose of parameters. We applied our model to infer the gene works from two
datasets, and essfully find all relationships that have been pre-set in the simulated data. The result
validates the feasibility and the efficiency of our model. From the real yeast expression data, our
method finds many significant candidate interactions between genes and a high percentage of which are
highly consistent with ones validated experimentally in public database.
Key words: gene work; probabilistic graphical model; factor graph; message passing
algorithm
0 引言
在后基因组时代,我们不仅要在分子水平上进一步挖掘基因信息,还要从基因组的角度
研究基因之间的相互作用机制,剖析生命体内的各种复杂系统[1]。基因表达数据不仅可以用
于分析基因表达规律,还能帮助研究基因间的调控关系,构建调控网络。基因调控网络是由
细胞中参与调控的 DNA、RNA、蛋白质及其代谢中间物所形成相互作用网络[2]。对基因调
控网