摘要
动态贝叶斯网络(DBN)是基因调控网络的一种有力建模工具。贝叶斯结构期望最大算法(SEM)能较好地处理构建基因调控网络中数据缺失的情况,但SEM算法学习的结果对初始参数设置依赖性强。针对此问题,提出一种改进的SEM算法,通过随机生成一些候选初始值,在经过一次迭代后得到的参数中选择一个最好的初始值作为模型的初始参数值,然后执行基本的SEM算法。利用啤酒酵母细胞周期微阵列表达数据,构建其基因调控网络并与现有文献比较,结果显示该算法进一步提高了调控网络构建的精度。
Dynamic Bayesian network (DBN) is a powerful moedling tool for gene rugulation network. Missing data in building gene regulation network is better dealt with SEM (Bayesion structure expectation maximization) algorithm, however, the result of learning by SEM algorithm has strong dependence on the initial parameters. This paper proposed an improver SEM algorithm, which randomly generated a number of candidate initial parameters and selected the best parameter as whole model' s initial parameter to execute basic SEM algorithm after a iterative process. Comparing gene regulation network constructed with yeast cycle gene expression data by improved SEM algorithm with existing literature, the result indicates further improve the accuracy of constructing regulation network.
出处
《计算机应用研究》
CSCD
北大核心
2010年第2期450-452,458,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(60705015)
安徽省自然科学基金资助项目(070412064)
合肥工业大学科学研究发展基金资助项目(070504F)
关键词
基因调控网络
动态贝叶斯网络
贝叶斯结构期望最大化算法
gene rugulation network
dynamic Bayesian network (DBN)
Bayesian structure expectation maximization algorithm