摘要
提出一种采用递归神经网络模型构建基因调控网络,将结构训练与参数训练相结合的方法进行网络的权值训练.采用模拟退火算法训练网络结构,找出调控关系权值,再引入基于免疫思想的粒子群算法对权值进行参数优化,得到基因调控网络图.并分别用人工数据和大肠杆茵DNA修复系统基因数据进行实验.实验结果表明,该方法能有效地从基因时序数据中揭示基因间的调控关系.
We constructed gene regulatory networks adopting recurrent neural network model. We proposed a two-step procedure for genetic regulatory network inference. At first we used simulated annealing algorithm to search network structure space and found meaningful weights that indicate the regulatory relations. Secondly we adopted improved particle swarm optimization algorithm based on immune principle to determine the network parameters. Our approach has been applied to both artificial data set and data set of Desoxyribonucleic acid （DNA） Repair System of Escherichia coll. The results demonstrate that the method can provide a meaningful insight into potential regulatory interactions between genes, which is revealed by the nonlinear dynamics of the gene expression time series. Thereby we have provided a new approach to solve the biological problem of constructing gene regulatory networks.
出处
《吉林大学学报（理学版）》
CAS
CSCD
北大核心
2010年第2期284-290,共7页
Journal of Jilin University:Science Edition
基金
国家自然科学基金(批准号:60673099
60873146
60973092)
国家高技术研究发展计划863项目基金(批准号:2007AA04Z114
2009AA02Z307)
吉林省生物识别新技术重点实验室项目(批准号:20082209)
关键词
基因调控网络
递归神经网络
模拟退火算法
免疫系统
粒子群算法
gene regulatory network
recurrent neural network
simulate annealing algorithm
immune system
particle swarm optimization
