基于多特征融合的基因调控网络构建方法研究  

Research on Gene Regulatory Networks Construction Based on Multi-featured Fusion

作　　者：孟军[1] 周广博 黄楚冰 

机构地区：[1]大连理工大学计算机科学与技术学院,辽宁大连116023

出　　处：《小型微型计算机系统》2016年第4期743-747,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61472061)资助;大连理工大学大学生创新创业训练计划项目(201410141065)资助

摘　　要：构建基因调控网络一直是生物信息学和系统生物学领域的一个研究热点.为了进一步提高基因调控网络构建的准确度,提出一种多特征融合的构建方法.分别对基因表达数据、序列信息和基因功能数据进行特征提取,利用特征级融合策略将三类特征融合,形成高维特征向量,经特征选择后得到最优的特征子集,进而利用支持向量机建立分类模型,同时考虑了样本不平衡对分类性能的影响.在拟南芥和番茄数据集上的实验结果表明该方法能够较好地预测出转录因子和目标基因之间的调控关系,获得准确的基因调控网络.Reliable inference of transcription regulatory networks is a challenging task in bioinformatics and system biology.In order to further enhance the accuracy of constructed gene regulatory networks,a method to construct gene regulatory networks based on multifeatured fusion is proposed.Features of gene expression data,sequence information and gene function data are extracted respectively,and then forms the higher-dimension feature vector by feature-level fusion strategy.A feature selection approach is used to rank the integrated feature vector and extract optimal features.Finally,SVM is applied for establishing a classification model based on these optimal features.M eanwhile,the impact of sample imbalance on the classification performance is studied.Experimental results on Arabidopsis thaliana dataset and tomato dataset showthat the proposed method can achieve an improved and robust performance in identifying true connections between TFs and their target genes,and thereby constructs gene regulatory networks accurately.

关 键 词：多特征融合 支持向量机 基因调控网络 特征选择 数据集成 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]