High-throughput RNA-sequencing technologies that provide spatial resolution of transcripts, popularly known as spatial transcriptomics, are on the rise. This technology seems highly promising and has an untapped potential for expression-driven discovery in development and disease. Nonetheless, it also faces the central challenge of mixed cell type signals due to limitations in resolution. This is apparent in sequencing-based 10X Visium where slides have larger spots of 55 um. This mixed transcriptional signal can pose inferential problems; however, it can theoretically be deconvoluted into underlying cell types. To this end, we developed a systematic deconvolution framework and performed benchmarking in previously unvalidated healthy and disease samples from human coronary arterial and kidney disease. We used: Cell2location, RCTD and spatialDWLS that have previously been shown to perform well in mouse brain and simulated data (1). We show that all three methods are capable of deconvoluting verifiable cell types when benchmarked against expert provided ground truth based on accuracy scores (0.7-0.73). Kidney podocyte cells and major populations of macrophages, smooth muscle cells and fibroblasts in arteries are all deconvoluted with a high level of agreement. Bayesian Cell2location is more computationally demanding, however it provides quality solutions, when less reference data is available.
Authors: Alban Obel Slabowska, Charles Pyke, Henning Hvid, Leon Eyrich Jessen, Simon Baumgart, Vivek Das
The datasets presented in this study can be found in online repositories as outlined below. Danish legislation regarding sharing of human data requires that the spatial transcriptomics data can be made available upon reasonable request to the corresponding author and approval from The Danish Data Protection Agency.
Atherosclerosis scRNA from Wirka et al. 2019, Pan et al. 2020, and Alsaigh et al. 2020. All of which can be obtained from the PlaqView Portal.
Kidney scRNA data was obtained from The Kidney Precision Medicine Project KPMP.
Spatial transcriptomics data was generated internally at Novo Nordisk using 10X Visium technology for FFPE samples.
RCTD was run using R 4.1.2 and the spacexr package (2.0.1). SpatialDWLS was run using the Giotto package (1.1.2) and R 4.1.2 with python 3.10 dependencies. Cell2location was run using the cell2location package (0.1) and python 3.10 and libraries pyro-ppl 1.8.0; scvi-tools 1.0.0; torch 1.9.0; numpy 1.23.4; scanpy 1.9.1; anndata 0.8.0. Packages Seurat 4.3.0, SeuratDisk 0.0.0.9020, SeuratData 0.2.2, tidyverse 1.3.2, ggplot 3.4.0 were used for data wrangling and visualizations.
Supplementary Figure 1. Single cell RNA reference data UMAP with cell type labels. Top are major CVD cell types, bottom are major cell types within CKD atlas. CKD abbreviations as follows: PT, Proximal Tubular; EC, Endothelial Cell; TAL, Thick Ascending Limb; ATL, Ascending Thin Limb; IC, Intercalated Cell; CNT,Connecting Tubule; DTL, Descending Thin Limb; DCT, Distal Convoluted Tubule; PEC, Parietal Epithelial Cell; POD, Podocyte; PC, Principal Cell.
Supplementary Figure 2: For each method deconvolution of sample CVD2 was repeated with multiple different reference sizes. Correlation between independent deconvolutions was calculated to determine degree of variation in prediction as a function of reference data size. Cell2location is observed to be more consistent across cell types.
Supplementary Figure 3: The RMSRD between each pair of methods, with a lower value indication more similar results for the specific cell type.
Supplementary Figure 4: Benchmarking the deconvolution time requirements across methods. For the reference preparation step the x-axis indicates number of cells in single cell reference dataset. For deconvolution, x-axis indicates number of spots subject to deconvolution.
