基于DAE的单细胞RNA测序数据聚类研究  

Research on single⁃cell RNA sequencing data clustering based on DAE

作　　者：何慧茹 李晓峰[1] 张鑫 柳楠[1] HE Huiru;LI Xiaofeng;ZHANG Xin;LIU Nan(School of Computer Science and Technology,Shandong Jianzhu University,Jinan 250101,China)

机构地区：[1]山东建筑大学计算机科学与技术学院,山东济南250101

出　　处：《现代电子技术》2020年第24期144-148,共5页Modern Electronics Technique

基　　金：国家自然科学基金(61902221);山东省自然科学基金(ZR2018MF012)。

摘　　要：传统数据降维方法处理单细胞RNA测序数据存在特征提取能力较差、聚类精度较低等问题,有必要引入深度学习方法以提高对复杂数据特征的提取能力。在对数据不进行任何人工筛选的条件下,利用DAE提取表达能力更强的数据特征,分别以K⁃means和DBSCAN聚类作为DAE的顶层设置形成DAE+K⁃means和DAE+DBSCAN组合模型,将这两种深度学习组合模型在Deng数据集上与传统聚类模型SC3进行对比。与SC3的0.73聚类精度相比,DAE+K⁃means和DAE+DBSCAN的聚类精度分别达到0.93和0.97,分别提高了0.2和0.24。实验结果表明,DAE在单细胞聚类领域具有广阔的应用前景。As the traditional data dimension reduction method in processing single⁃cell RNA⁃sequencing data has some problems,such as poor feature extraction ability and low clustering accuracy,it is necessary to introduce the deep learning method to improve the extraction ability of complex data feature.Without any manual screening of data,the deep auto⁃encoder(DAE)is used to extract data feature with stronger expression ability.The K⁃Means and DBSCAN clustering is taken as the top⁃layer setting of DAE respectively to form DAE+K⁃Means or DAE+DBSCAN combined model,and the two deep learning combined models are compared with the traditional clustering model SC3 on Deng dataset.In comparison with the 0.73 clustering accuracy of SC3,the clustering accuracy of DAE+K⁃Means and DAE+DBSCAN reaches 0.93 and 0.97,respectively,which is improved by 0.2 and 0.24,respectively.The experimental results show that the DAE has a broad application prospect in the field of single⁃cell clustering.

关 键 词：单细胞聚类 深度自动编码器 深度学习 K⁃means聚类 DBSCAN聚类 结果分析 

分 类 号：TN919-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]