spatialLIBD: an R/Bioconductor package to visualize spatially-resolved transcriptomics data

1The rise of spatially-resolved transcriptomics technologies is providing new opportunities to answer questions about the structure and function of complex intact tissues. However, this exciting opportunity requires the development of new computational methods and interactive software. These tools have to accommodate the analysis of transcriptomic data with spatial coordinates and high resolution images into previously developed methods for single cell RNA-sequencing (scRNA-seq) analysis, which lacked these components.

2Currently there are limited options for interactive exploration of spatially-resolved transcriptomics data. The 10x Genomics Loupe Browser utilizes a file created by the Space Ranger data pre-processing pipeline and performs marker gene analysis, as well as unsupervised graph-based and k-means clustering. Loupe visualizes both the spatially-resolved transcriptomics data along with the high resolution imaging data to export gene expression maps. An alternative to Loupe is the Giotto pipeline, which contains both a data analysis and data visualization modules. The data analysis module has many functions ranging from pre-processing to more advanced analysis such as characterizing cell-cell interactions. The visualization module contains an interactive workspace for exploring multiple layers of information. However, Loupe and Giotto currently do not support visualizing more than one tissue section at a time, which is useful for comparing replicates and annotating observed spots on the Visium array across samples. Seurat provides guidelines for processing and visualizing spatially-resolved transcriptomics data at https://satijalab.org/seurat/articles/spatial_vignette.html. Seurat does not however provide interactive visualizations or the ability to create an interactive website. Furthermore, while unsupervised clustering methods are widely developed, manual annotation of spots using known marker genes is important, as well as cross-sample dimension reduction techniques such as UMAP and t-SNE.

3spatialLIBD is an R/Bioconductor package that provides functions to import output from Space Ranger including clustering (k-means, graph-based) and dimension reduction results (PCA, UMAP, t-SNE), interactively visualize and explore spatially-resolved transcriptomics data, and provides functionality to manually annotate spots. It was developed initially for visualizing multiple Visium tissue sections from the human brain, yet it is flexible enough to be used with other Visium data from different tissue types (Supplementary File 1, Fig. 1A).

4The primary functions in spatialLIBD allow users (i) to visualize the spot-level spatial gene expression data or any other continuous variable, and data-driven clusters or any categorical variable, with the histology image in the background (Fig. 1B), and (ii) to inspect the data interactively, either on the user’s computer or through web hosting services such as shinyapps.io (Fig. 1C). In addition, spatialLIBD enables users to visualize multiple samples at the same time, to export all static visualizations as PDF files or all interactive visualizations as PNG files, as well as all result tables as CSV files. Additional file 1 shows how to read in data processed with Space Ranger into R, make plots or create an interactive website with spatialLIBD.

5Additional file 2 illustrates how spatialLIBD can be used to visualize clustering results derived from multiple samples as well as other continuous covariates, using a human brain dataset with twelve samples. In particular, clusters derived from all samples facilitate their interpretation since cluster numbers (and thus their meaning) will be the same across all samples. The companion interactive website for this dataset is available at http://spatial.libd.org/spatialLIBD. Additional file 1 shows the steps necessary to make your own interactive website, such as https://libd.shinyapps.io/spatialLIBD_Human_Lymph_Node_10x/.

6spatialLIBD has evolved over time and now provides features that further facilitate the visual inspection of the results. These include nine color scales, including colorblind-friendly scales as well as the option to reverse the color scale gradient order (Fig. 2A). The transparency of the spots can be controlled as well. Multiple images are supported and can be edited on the web application to increase brightness or perform other image manipulations that can reveal hidden patterns to the eye (Fig. 2B). Side by side plots of the images and the clusters or gene expression data can also be generated.

7spatialLIBD is designed to work with SpatialExperiment R/Bioconductor objects. It uses shiny and plotly for the interactive website, ggplot2 and cowplot for the static images. Interactive image editing is powered by magick. Colorblind-friendly colors are provided by viridis.

8spatialLIBD benefits from the open-source Bioconductor ecosystem and relies on the infrastructure provided by SpatialExperiment. It provides user-friendly functionality for visualizing continuous and categorical measurements along with the tissue images from spatially-resolved transcriptomics data from projects involving one or more samples measured on the 10x Genomics Visium platform. Since spatialLIBD is compatible with SpatialExperiment, spatialLIBD could be used for other spatially-resolved transcriptomic platforms as they become better supported by SpatialExperiment.

9spatialLIBD provides a simple interactive website, which can be used for sharing data on the web, as well as manually annotating spots. spatialLIBD has limitations (Additional file 1) that are inherent to the methods used to implement it, such as (i) the memory per user required by a server for hosting the web application, (ii) response speeds for the interactive views due to the number of spots, (iii) the resolution of the images displayed limiting the usefulness to magnify specific spots, and (iv) customization of the web application by the end user. Despite these limitations, open-source software solutions like spatialLIBD enable faster data exploration and insights in spatially-resolved transcriptomics research projects and can serve as a testing ground for ideas.

10While some visualization software alternatives do not provide visualizations across multiple samples, static visualizations of multiple samples can be combined using R packages such as cowplot or patchwork or interactive versions using plotly, which is how spatialLIBD was implemented. As spatially-resolved transcriptomics becomes more widely available, new methods are being implemented in other languages, such as Python. One such example is squidpy, which is currently limited to one Visium sample as noted at https://github.com/theislab/squidpy/issues/417. squidpy does provide exciting and powerful features, and bridges the R and Python worlds.

11spatialLIBD is fully integrated with the R/Bioconductor open source ecosystem and is available at https://bioconductor.org/packages/spatialLIBD. It provides user-friendly functionality for importing results from the 10x Genomics Space Ranger pipeline for processing spatially-resolved transcriptomics data from Visium. spatialLIBD provides different options for visualizing and interactively exploring the data from multiple Visium capture areas, and can create websites that can be self-hosted such as http://spatial.libd.org/spatialLIBD/ or hosted at shinyapps.io such as the example website https://libd.shinyapps.io/spatialLIBD_Human_Lymph_Node_10x/.

12Project name:spatialLIBD: an R/Bioconductor package to visualize spatially-resolved transcriptomics data

13Project home page: https://bioconductor.org/packages/spatialLIBD

14Operating system(s): Platform independent

15Programming language: R

16Other requirements: Bioconductor version 3.13 or newer to access the latest features

17License: Artistic-2.0

18Any restrictions to use by non-academics: none