| 引用本文: | 刘腾,李鑫,印明柱.基于变分自编码器的空间转录组细胞聚类研究[J].生物信息学,2024,22(4):270-276. |
| LIU Teng,LI Xin,YIN Mingzhu.Variational autoencoder enabled cell clustering method for spatial transcriptomics[J].Chinese Journal of Bioinformatics,2024,22(4):270-276. |
|
| 摘要: |
| 空间转录组测序技术能够在生成基因表达谱的同时,保留细胞在组织内部的位置信息。如何充分利用基因表达谱和空间位置信息来识别空间区域,完成细胞亚群聚类是空间转录组学数据分析的基础和关键。本文提出基于变分自编码器和图神经网络结合的空间转录组细胞亚群聚类方法。构建双层编码器结构,每一层包含简化图卷积(Simple graph convolution, SGC),用以生成低维表征。解码器用以重构特征矩阵,通过最小化损失函数来提高低维表征质量。对低维表征进行下游聚类,生成不同的细胞亚群。本文提出的聚类方法与多个基准方法在常用的空间转录数据集上进行比较,在聚类准确性和适应性方面都有优势,证明了该方法的有效性。 |
| 关键词: 空间转录组学 变分自编码器 图神经网络 细胞聚类 |
| DOI：10.12113/202305003 |
| 分类号:Q2 |
| 文献标识码:A |
| 基金项目:科技部重点研发基金项目(No. 2022YFC3601800); 重庆市教育委员会科学技术研究计划重大项目(No. KJZD-K202300105); 重庆大学附属三峡医院基础医学重点项目(No. 2023YJKYXML-001). |
|
| Variational autoencoder enabled cell clustering method for spatial transcriptomics |
|
LIU Teng, LI Xin, YIN Mingzhu
|
|
(Chongqing University Three Gorges Hospital, Wanzhou 404000, Chongqing, China)
|
| Abstract: |
| Spatial resolved transcriptomics technology can simultaneously generate gene expression profiles while preserving the positional information of cells within the tissue. How to fully utilize gene expression profiles and spatial positional information to identify spatial regions and complete cell subpopulation clustering is the basis and key for spatial transcriptomics data analysis. In this paper, a spatial transcriptomics cell clustering method based on the combination of variational autoencoder and graph neural network is presented. A two-layer encoder structure is constructed, with each layer containing Simple graph convolution (SGC) to generate low-dimensional representations. The decoder is used to reconstruct the feature matrix and improve the quality of low-dimensional representations by minimizing the loss function. Downstream clustering is performed on the low-dimensional representations to generate different cell subpopulations. The proposed clustering method is compared with several benchmark methods on multiple datasets and has advantages in clustering accuracy and adaptability, demonstrating the effectiveness of the proposed method. |
| Key words: Spatial transcriptomics Variational autoencoder Graph neural network Cell clustering |
