基于引力场算法的基因调控网络构建  

Reconstruction of gene regulatory network based on gravitation field algorithm

作　　者：郑明[1] 刘桂霞[1] 周柚[1] 周春光[1] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012

出　　处：《吉林大学学报（工学版）》2014年第2期427-432,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(60973092;60903097;61175023);吉林大学'985工程'项目;吉林大学研究生创新项目(20121109)

摘　　要：为了解决传统基因调控网络构建算法准确度不高且效率低下的问题,使用一种基于微分方程的新型网络构建算法。算法分为奇异值分解和引力场算法两部分,奇异值分解策略用来缩小解空间范围,提高运行效率。引力场算法是本文核心,共分初始化、解空间分解、移动算子和吸收算子4步骤。分解策略采用随机分组法,移动算子采用元素逐个移动法,并可根据收敛效果重新移动。最后,将本文算法与另两种启发式搜索算法下的网络构建进行比较,构建模拟和真实的基因调控网络。实验结果显示:本文算法具有更高的执行效率。In order to resolve the low accuracy and inefficiency of reconstruction of Gene Regulatory Networks （GRNs） in system biology, we proposed a novel inference algorithm from gene expression data based on differential equation model. In this algorithm, two methods are employed for inferring GRNs. One is Singular Value Decomposition （SVD） method and the other one is Gravitation Field Algorithm （GFA）. The SVD method is used to decompose gene expression data, determine the algorithm solution space, and get all candidate solutions of GRNs. The GFA is the kernel part of the proposed algorithm. The GFA is divided into four parts： initialization, division of solution space, movement operator and absorption. Random group method is used in division of solution space. Every element movement method is used in movement operator. The proposed algorithm is validated on both the simulated scale-free network and real benchmark gene regulatory network in network database. Both genetic algorithm and simulated annealing are also used to evaluate GFA. The cross-validation results confirm the effectiveness of the proposed existing algorithms. algorithm, which outperforms significantly other

关 键 词：人工智能 引力场算法 基因调控网络 优化算法 奇异值分解 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]