《生命科学》 2023, 35(6): 833-842

图神经网络在单细胞测序数据上的研究进展

王红麟¹ , 曲　红^2,* , 赵　敏^3,* , 曲大成^1,4,*

¹北京理工大学计算机学院，北京 100081 ²北京大学生命科学学院，北京 100871 ³澳大利亚阳光 海岸大学科学与工程学院，马卢奇郡，昆士兰州 4558 ⁴中国科协信息中心，北京 100863

摘　要：

单细胞测序技术使得科研人员能够以细胞级别的分辨率进行基因表达数据分析，以此发现组织( 如肿瘤组织或器官组织) 中具有异质性的细胞。这项技术对癌症病理学的研究、生命发育过程的探索等起到了重大推动作用。单细胞测序数据有着样本量大、特征多且稀疏的特点，因此近些年一些研究工作尝试使用图神经网络进行单细胞测序数据的挖掘。这些研究工作一般先根据细胞内的基因表达信息将单细胞测序数据转化为细胞图结构，然后使用图神经网络聚合细胞间邻域信息来进行细胞表示学习，并在细胞聚类任务和细胞类型标注任务上取得了很好的效果。本文旨在介绍图神经网络在单细胞测序数据挖掘上的研究进展，并设计实验展示scGNN、scGCN 和scDeepSort 三个主流的用于单细胞测序数据的图神经网络模型的性能。最后，结合研究进展与实验分析，本文对图神经网络处理单细胞测序数据这一领域的未来研究方向进行了展望，以促进图神经网络更好地服务于单细胞测序数据的挖掘。

通讯作者：曲　红 , Email:quh@mail.cbi.pku.edu.cn 赵　敏 , Email:mzhao@usc.edu.au 曲大成 , Email:qudc@bit.edu.cn

Research progress of graph neural network on scRNA-seq

WANG Hong-Lin¹ , QU Hong^2,* , ZHAO Min^3,* , QU Da-Cheng^1,4,*

¹School of Computer Science & Technology, Beijing Institute of Technology, Beijing 100081, China ²College of Life Sciences, Peking University, Beijing 100871, China ³School of Science and Engineering, University of the Sunshine Coast, Maroochydore DC, QLD 4558, Australia ⁴Information Center, China Association for Science and Technology, Beijing 100863, China

Abstract:

Single-cell RNA sequencing (scRNA-Seq) technology enables researchers to analyze gene expression data at single-cell resolution to discover heterogeneous cells in tumor tissues or organ tissues, which has played a major role in the study of cancer pathology and the exploration of life development processes. scRNA-Seq data usually contain a large number of samples with numerous features and sparsity. Therefore, in recent years, some research efforts have attempted to mine scRNA-Seq data using graph neural networks. These graph neural network methods typically first transform scRNA-Seq data into cell graph structures based on intracellular gene expression information, and then aggregate neighborhood information among cells for cell representation learning. They have achieved good results in cell clustering and cell type labeling. This paper aims to introduce the current research progress of graph neural network methods in scRNA-Seq data mining. We select three graph neural network models, scGNN, scGCN and scDeepSort, and compare their performance in scRNA-Seq data analysis. Finally, combined with the research progress and experimental results, we propose the future research prospects for graph neural network in scRNA-Seq data mining.

Communication Author：QU Hong , Email:quh@mail.cbi.pku.edu.cn ZHAO Min , Email:mzhao@usc.edu.au QU Da-Cheng , Email:qudc@bit.edu.cn