Mobile Menu

Updated computational tools for spatial transcriptomics analysis

Spatial molecular analysis is not new. However, many subcellular, high-multiplex and multi-omics spatial methods have only emerged over the past few years. These spatial transcriptomics technologies offer an unparalleled view of molecular biology. A continual challenge for researchers in the field is interacting with and handling this high-dimensional spatial data.

Dr. Jinmiao Chen, Principal Investigator, Agency for Science, Technology and Research, is the primary developer of several new tools for working with spatial data and recently spoke at a Front Line Genomics webinar. In this article, we will give an overview of her recent computational tools and the utility they might have for you. Please note, this piece uses quotes taken directly from the webinar. To hear the full set of webinar talks on Gaining New Insights into Biology from Spatial Data Analysis Approaches, please follow the webinar link.

STAMP-ing down dimensionality

High dimensional spatial transcriptomics is ideal for studying tissue complexity. Naturally, the analysis of spatial data should always try to take advantage of the spatial dimension. This means that single-cell methods are no longer optimal for spatial data. New spatial tools are needed.

Jinmiao: “In my lab, we create a series of AI tools for the analysis of spatial omics. STAMP and SEDR for dimension reduction. GraphST for multi-sample integration and cell type deconvolution and lastly, we have SpatialGlue, which was recently for spatial multiomics [integration]”

The first tool considered here is STAMP1, which is designed for interpretable spatially-aware dimension reduction of spatial transcriptomics. It takes spatial transcriptomic data and outputs a biologically-meaningful low dimensional representation of the data. This enables meaningful interpretation and downstream analysis. It can also perform multi-sample analysis by projecting numerous samples into the same low-dimensional space.

GraphST – clusters, deconvolutes and integrates

The next tool considered here is GraphST2, which is a graph neural network model with self-supervised learning. It is designed to perform multiple tasks in spatial analysis, namely, spatially informed clustering, multi-sample integration and cell-type deconvolution.

For clustering cells and spatial regions, GraphST outperformed current methods such as STAGATE, Giotto and Seurat in clustering 10x Visium human brain data into distinct brain layers. It also revealed finer-grained tissue structures in Stereo-seq mouse embryo data compared to the original publication3 (Figure 1).

The second task it can perform is multi-sample integration to overcome batch effects in spatial data between sections by jointly projecting data from sections into the same spatial domains.

The final task that GraphST can perform is cell type deconvolution, which is achieved by projecting pre-existing single-cell atlas data onto spatial transcriptomics in order to automatically label cells.

Figure 1. GraphST outperforms other methods in spatial clustering. Image taken directly from FLG webinar (September 19th, 2023), full credit – Dr. Jinmiao Chen, adapted from Long, et al. 2

SpatialGlue-ing Omics together

Finally we have SpatialGlue4. It was developed for spatial multi-omics, and it is a graph neural network with a dual attention mechanism. Typical spatial multi-omics methods profile RNA & Protein (Cite-seq) or RNA and epigenomics (ATAC-seq or CUT&Tag) and SpatialGlue can handle either.

Jinmiao: “With the first attention, we integrate spatial information with the omics measurement within each data modality. Then, with the second attention, we integrate across modalities, such as, integrating RNA with protein or integrating RNA with ATAC-seq.”

When testing SpatialGlue with simulation data, it performed better than other methods such as MultiVI and Seurat. When it was tested with existing spatial multi-omics data, such as the epigenome-transcriptome spatial data from Zhang, et al. 5, SpatialGlue allowed a clear separation of different tissue regions using the two omics together, better than with each omic individually (Figure 2).

Figure 2. SpatialGlue merges transcriptomic and epigenomic data to gather new insights. Image taken directly from FLG webinar (September 19th, 2023), full credit – Dr. Jinmiao Chen.

Webinar Q&A’s (summarised – not direct quotes)

Q: Does SpatialGlue theoretically work with any multi-omics combination? Could it handle three or more omics?

A: Currently SpatialGlue only works with two modalities at once and the specific ones listed. But the model is being expanded to handle 3 or more omics. The GitHub will be updated when this happens.

Q: Can SpatialGlue be used to integrate two omics modalities acquired separately from two neighbouring sections of the same tissue?

A: Currently it only integrates omic data from the same tissue. Another algorithm is under development that allows integration across tissue sections

Q: What would you like to happen to make everyone’s life easier when analysing spatial data?

A: Cell segmentation is a critical part, if your cells are segmented wrong, all the downstream analysis is wrong. And with the latest subcellular spatial data, some of them don’t have the high quality of image data needed to segment. Certain tissues also have such crowded cells that you cannot manually segment the cells.

The second challenge is the large data size. With the latest developments, we can acquire data from multiple samples, across the whole genome, at subcellular resolution. The data is getting larger and larger and so we are more dependent on GPU, CPU, etc. Computational speed is another issue we need to address.

To hear from the other speakers at this webinar and for the full Q&A please use this link.


1.            Chen, J., Zhong, C. & Ang, K.S. Interpretable spatially aware dimension reduction of spatial transcriptomics with STAMP. (2023).

2.            Long, Y. et al. Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST. Nature Communications 14, 1155 (2023).

3.            Chen, A. et al. Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays. Cell 185, 1777-1792. e21 (2022).

4.            Long, Y. et al. Integrated analysis of spatial multi-omics with SpatialGlue. bioRxiv, 2023.04.26.538404 (2023).

5.            Zhang, D. et al. Spatial epigenome–transcriptome co-profiling of mammalian tissues. Nature 616, 113-122 (2023).