基于语义特征向量的DNA与转录因子结合特异性预测  

BINDING SPECIFIC PREDICTION OF DNA AND TRANSCRIPTION FACTOR BASED ON SEMANTIC FEATURE VECTOR

作　　者：孙晓雨 权丽君[1,2] 梅杰 黄立群 吕强 Sun Xiaoyu;Quan Lijun;Mei Jie;Huang Liqun;LüQiang(School of Computer Science and Technology,Soochow University,Suzhou 215006,Jiangsu,China;Jiangsu Province Key Lab for Information Processing Technologies,Suzhou 215006,Jiangsu,China)

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006 [2]江苏省计算机信息处理技术重点实验室,江苏苏州215006

出　　处：《计算机应用与软件》2022年第2期38-43,112,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(31801108);苏州大学江苏省计算机信息处理重点实验室开放课题(KJS1843);江苏高校优势学科建设工程资助项目。

摘　　要：基因表达的生物系统受到DNA和转录因子(Transcription Factor,TF)相互作用的调控,但是对于DNA和TF的结合机制,人们仍然不够了解。为此,提出一个结合深度学习和SVC的模型预测TF与DNA序列结合特异性,并将这一方法命名为semanticSVC。在ENCODE项目中的多个TF实验数据集的集合上进行深度学习模型训练,从而挖掘出跨越多个TF的全局语义特征。基于这些语义特征通过浅层学习模型SVC快速对目标任务构建预测模型,进行预测结果的可视化分析。与现有方法相比,该方法在预测结合特异性方面取得了更优的性能。The biological system of gene expression is regulated by the interactions between DNA sequences and transcription factors(TF),but people do not have enough knowledge about their binding mechanism.In order to solve such problems,this paper proposes a model combining deep learning method and SVC to predict the binding specificities of DNA sequences and TF,and named it as semanticSVC.The deep learning model was trained on the set of multiple TFs experimental datasets in the ENCODE project,so as to mine the global semantic features across multiple TFs.Based on these semantic features,the target task prediction model was quickly built through the shallow learning model SVC.The visual analysis of the prediction results was carried out.Compared with the existing methods,this method has better performance in predicting binding specificity.

关 键 词：DNA特异性 转录因子 深度学习 SVC t-SNE降维 特征向量可视化 单点突变 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]