单细胞RNA测序数据插补方法综述  

Review of imputation methods based on single-cell RNA-sequencing

作　　者：张少强[1] 李荔瑄 谢林娟 吕庆 ZHANG Shaoqiang;LI Lixuan;XIE Linjuan;LYU Qing(College of Computer and Information Engineering,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学计算机与信息工程学院,天津300387

出　　处：《天津师范大学学报（自然科学版）》2023年第2期1-10,73,共11页Journal of Tianjin Normal University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61572358);天津市应用基础与前沿技术研究计划重点资助项目(19JCZDJC35100);天津市自然科学基金青年基金资助项目(18JCQNJC74100).

摘　　要：单细胞RNA测序(scRNA-seq)数据插补方法用于解决scRNA-seq数据观测中存在的大量“漏失”(dropout)噪音,改善下游分析,scRNA-seq数据插补方法设计是单细胞数据研究的热点方向之一.本文首先对20种主要的scRNA-seq数据插补方法进行介绍,包括基于模型的插补方法(6种)、基于平滑的插补方法(3种)、基于深度学习的插补方法(8种)和基于低秩矩阵的插补方法(3种),分析了各类方法的优势和缺点;其次,简要综述了插补方法比较研究的相关成果;然后,针对4种下游数据分析评估了以上方法(除scGNN外)的性能;最后,分析目前scRNA-seq插补所面临的挑战,并指出新的研究方向.Imputation method for single-cell RNA sequencing(scRNA-seq)data is used for solving large numbers of dropout noise observed in scRNA-seq data and improving the downstream analysis.Imputation method design for scRNA-seq data is one of the hot research directions of single cell data.Firstly,20 kinds of main imputation methods are introduced,including model based imputation methods(six kinds),smoothing based imputation methods(three kinds),deep learning based imputation me-thods(eight kinds)and low rank matrix based imputation methods(three kinds),and the advantages and disadvantages of each class of method are analyzed.Then the results of comparative research on imputation methods are briefly reviewed.And then the performances of the above methods(except for scGNN)are evaluated by four kinds of downstream data analysis.Finally,the challenges faced by current scRNA-seq imputation methods are analyzed,and new research directions are pointed out.

关 键 词：单细胞RNA测序 插补 细胞类型 差异表达 轨迹推断 

分 类 号：Q811.4[生物学—生物工程]