基于特征图网络和多种生物信息预测关键蛋白质的深度学习框架  

Deep Learning Framework for Predicting Essential Proteins Based on Feature Graph Network and Multiple Biological Information

作　　者：刘桂霞[1] 曹心恬 赵贺 LIU Guixia;CAO Xintian;ZHAO He(Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education,College of Computer Science and Technology,Jilin University,Changchun 130012,China)

机构地区：[1]吉林大学计算机科学与技术学院,符号计算与知识工程教育部重点实验室,长春130012

出　　处：《吉林大学学报（理学版）》2024年第3期593-605,共13页Journal of Jilin University:Science Edition

基　　金：国家自然科学基金(批准号:62372208,61772226);吉林省科技发展规划重点项目(批准号:20210204133YY).

摘　　要：针对生物实验识别关键蛋白质费时费力,使用计算方法预测关键蛋白质无法有效整合生物信息的问题,提出一个深度学习框架.首先利用网络拓扑结构、基因表达数据和GO(gene ontology)注释数据构建加权蛋白质相互作用网络;然后分别使用特征图网络和双向长短期记忆细胞从亚细胞定位数据、蛋白质复合物数据和基因表达数据中提取特征向量;最后将这些特征向量输入到任务学习层预测关键蛋白质.实验结果表明,相比于现有的计算方法,该方法预测性能更好.Aiming at the problem that identifying essential proteins in biological experiments was time-consuming and laborious,and using computational methods to predict essential proteins could not effectively integrate biological information,we proposed a deep learning framework.Firstly,a weighted protein interaction network was constructed by using network topology structure,gene expression data and gene ontology(GO)annotated data.Secondly,feature vectors were extracted from subcellular localization data,protein complex data and gene expression data by using feature graph network and bi-directional long short-term memory cells,respectively.Finally,these feature vectors were input into the task learning layer to predict essential proteins.The experimental results show that,compared with existing computational methods,the proposed method has better predictive performance.

关 键 词：关键蛋白质 特征图网络 亚细胞定位 基因表达 GO注释 蛋白质复合物 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]