整合单细胞转录组数据的空间转录组解卷积

Spatial Transcriptome Deconvolution Integrating Single-Cell Transcriptome Data

摘要: 空间转录组技术是理解组织结构和功能的关键。然而，许多空间转录组技术没有单细胞分辨率，其基因组数据反映了细胞群的平均表达水平，在此基础上的差异、聚类和关联分析等数据分析问题实际是在细胞群水平上进行的，忽略了不同细胞类型之间的差异，导致分析结果出现偏差甚至错误。本文开发了一种解卷积算法stDRPC，该算法首先从单细胞转录组数据中得到细胞类型特异性表达信息矩阵，将该矩阵作为参考基矩阵，然后利用稳健偏相关性算法对空间转录组数据进行基于参考基矩阵的解卷积，从而得到各细胞类型所占比例。小鼠脑皮层及人胰腺导管腺癌(PDAC)数据分析表明所提算法在准确性及生物学意义上都优于常用算法。

Abstract: Spatial transcriptome technology is crucial for understanding tissue structure and function. However, many spatial transcriptome techniques do not have single-cell resolution, and their genomic data reflects the average expression level of cell populations. On this basis, the data analysis problems such as differential, clustering and association analysis are actually carried out at the cell population level, ignoring the differences among different cell types, resulting in biases and errors. This article develops a deconvolution algorithm called stDRPC, which first obtains cell type specific expression information matrix from single-cell transcriptome data, uses this matrix as a reference-based matrix, and then uses a robust partial correlation algorithm to deconvolve the spatial transcriptome data based on the reference-based matrix to obtain cell type proportions. Data analyses of mouse cortical and human pancreatic ductal adenocarcinoma (PDAC) showed that the proposed algorithm is superior to commonly used algorithms in terms of accuracy and biological significance.