Perspective

Advances in spatial transcriptomic data analysis

Ruben Dries,1,2,3 Jiaji Chen,1 Natalie del Rossi,4 Mohammed Muzamil Khan,1,2,3
Adriana Sistig,4 and Guo-Cheng Yuan4,5
1
Department of Medicine, Boston University School of Medicine, Boston, Massachusetts 02118, USA; 2Bioinformatics Graduate
Program, Boston University, Boston, Massachusetts 02215, USA; 3Section of Computational Biomedicine, Boston University School of
Medicine, Boston, Massachusetts 02118, USA; 4Department of Genetics and Genomic Sciences, Charles Bronfman Institute for
Personalized Medicine, Icahn School of Medicine at Mount Sinai, New York, New York 10029, USA; 5Precision Immunology Institute,
Icahn School of Medicine at Mount Sinai, New York, New York 10029, USA

Spatial transcriptomics is a rapidly growing field that promises to comprehensively characterize tissue organization and ar-
chitecture at the single-cell or subcellular resolution. Such information provides a solid foundation for mechanistic under-
standing of many biological processes in both health and disease that cannot be obtained by using traditional technologies.
The development of computational methods plays important roles in extracting biological signals from raw data. Various
approaches have been developed to overcome technology-specific limitations such as spatial resolution, gene coverage, sen-
sitivity, and technical biases. Downstream analysis tools formulate spatial organization and cell–cell communications as
quantifiable properties, and provide algorithms to derive such properties. Integrative pipelines further assemble multiple
tools in one package, allowing biologists to conveniently analyze data from beginning to end. In this review, we summarize
the state of the art of spatial transcriptomic data analysis methods and pipelines, and discuss how they operate on different
technological platforms.

Multicellular organisms consist of tissues and organs, each special- like putting together a complex jigsaw puzzle from individual piec-
izing in a subset of biological processes and performed by the coor- es, the precise position and organization of the cells matter. The
dinated activities of many cells. Although all normal cells share the tissue environment plays a critical role during development in
same genome, their gene expression patterns and morphology can which, for example, it defines asymmetric cell fate decisions and
be drastically different. This variation is caused not only by inter- instructs cell movement. Positional information continues to be
nal gene regulatory circuitry differences but also by signaling from crucial at the adult stage to exert tissue-specific functions, to main-
the external tissue environment. Whereas decades of genome- tain tissue homeostasis, and to respond to external cues or pertur-
wide studies have accumulated large amounts of information bations. Notably, in diseases such as cancer, the normal tissue
about cell type–specific gene regulatory circuitries, our under- environment can be reprogrammed and manipulated to promote
standing of the external cell–tissue environment interactions re- malignant cell expansion, which is normally suppressed (White-
mains limited. side 2008), whereas deep understanding of the tumor immune
Recent years have witnessed an explosion of technological environment is essential for developing effective immunothera-
advances that collectively enable system-level characterization of peutic approaches (Binnewies et al. 2018).
cellular heterogeneity and spatial organization of tissues/organs. During the past few years, various technologies have been de-
Perhaps most notably is the rapid development of single-cell veloped for transcriptomic profiling while preserving spatial infor-
RNA-seq technology (scRNA-seq) applications that made it possi- mation. Collectively, these technologies have been named as the
ble to profile and compare the gene expression patterns of a large method of the year of 2020 by Nature Methods (Marx 2021) to rec-
number of individual cells within a tissue/organ (Svensson et al. ognize their importance and are expected to rapidly transform bi-
2018b). Together with the development of a rich set of computa- ological research in the coming years. Currently, there exist three
tional methods for data analysis (for review, see Yuan 2019; Hie major approaches that are engaged to spatially explore large pieces
et al. 2020), the scRNA-seq field has fulfilled a key role in the dis- of tissue and aim to perform this at single-cell resolution and on a
covery of novel cell types and laid the foundation for the creation genome-wide scale. First, sequential fluorescent in situ hybridiza-
of comprehensive cell atlases in different species (Han et al. 2018, tion (FISH)–based methods use a targeted approach, which is based
2020; Sebé-Pedrós et al. 2018; Spanjaard et al. 2018; Tabula Muris on predesigned probes. By introducing clever barcoding strategies
Consortium et al. 2018; Cao et al. 2019; Packer et al. 2019; Pijuan- combined with sequential hybridization and imaging, they can
Sala et al. 2019). However, a key step in the experimental process is identify the exact position of tens to thousands of individual tran-
the creation of a single-cell suspension through mechanical and scripts within a fixed tissue specimen (Lubeck et al. 2014; Chen
enzymatic dissociation steps, which inherently destroys the origi- et al. 2015; Shah et al. 2016; Codeluppi et al. 2018; Moffitt et al.
nal tissue architecture. 2018; Eng et al. 2019; Kishi et al. 2019; Goh et al. 2020). Second,
As such, reconstructing the structure of a tissue from its cellu- spatial labeling technologies use ingenious ways to link all tran-
lar components alone is extremely difficult, if not impossible. Just scripts within a spatial unit with known coordinates. This in situ
capturing step is subsequently followed by an unbiased standard
sequencing approach (Ståhl et al. 2016; Rodriques et al. 2019;

Corresponding authors: [email protected], [email protected]

Article and publication date are at https://www.genome.org/cgi/doi/10.1101/ © 2021 Dries et al. This article, published in Genome Research, is available un-
gr.275224.121. Freely available online through the Genome Research Open der a Creative Commons License (Attribution 4.0 International), as described at
Access option http://creativecommons.org/licenses/by/4.0/.

1706 Genome Research 31:1706–1718 Published by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/21; www.genome.org
www.genome.org
 Spatial transcriptomic data analysis

Vickovic et al. 2019; Liu et al. 2020; Merritt et al. 2020; Chen et al. tions results in a clear visual improvement of the vertical align-
2021; Cho et al. 2021; Stickels et al. 2021). Third, select genes can ment of the spatial expression data (Fig. 2B).
be targeted for in situ sequencing (ISS), with the synthesized cDNA Both FISH- and ISS-based techniques provide single-cell or
products labeled by fluorescent nucleotides and detected by imag- even subcellular resolution; however, this depends on proper iden-
ing (Lee et al. 2014; Wang et al. 2018; Qian et al. 2020; Hu et al. tification of cell morphology. Tracing the cell boundaries—and
2020b; Alon et al. 2021; Fu et al. 2021). For more detailed informa- other cellular structures such as the nucleus—is often referred to
tion about the different spatial units and the capturing and linking as cell segmentation (Fig. 2C). Although cell segmentation may ap-
strategies, the reader is referred to spatial technology reviews (Asp pear rather simple to human eyes, it has been proven hard to auto-
et al. 2020; Liao et al. 2021). Here we use four data sets to illustrate mate. Segmentation difficulties are further aggravated by factors
the outcome or methodology of different spatial transcriptomic such as cell density (e.g., solid tumors) or complex cell shapes
analysis steps: (1) a genome-wide spatial transcriptomics data set (e.g., neurons). A large number of methods have already been de-
generated from multiple slices of a breast tumor biospecimen veloped with gradual improvements in the accuracy and quality
(Andersson et al. 2020b), (2) a subcellular spatial data set from a (for reviews, see Dimopoulos et al. 2014; Vicar et al. 2019). More re-
whole-mouse coronal brain slice with approximately 500 genes cently, with the surge in deep learning frameworks and applica-
across about 78,000 generated by VIZGEN with MERFISH technol- tions, there have been some considerable improvements in the
ogy (https://info.vizgen.com/mouse-brain-data), (3) a genome- creation of generalizable cell segmentation and image registration
wide spatial data set from the human heart created by the Visium tools (Schmidt et al. 2018; Berg et al. 2019; Falk et al. 2019; Perkel
platform from 10x Genomics (https://support.10xgenomics.com/ 2019; Greenwald et al. 2021; Stringer et al. 2021). Finally, each
spatial-gene-expression/datasets/1.1.0/V1_Human_ Heart), and spot needs to be identified and uniquely assigned to a gene. This de-
(4) another subcellular spatial data set covering 10,000 genes coding strategy is typically intertwined with the technological
from hundreds of cells within the mouse somatosensory cortex setup and design, but there are efforts to make this more generaliz-
and generated by the seqFISH+ technology (Fig. 1; Eng et al. 2019). able and available to the broad community (Fig. 2C; Perkel 2019).
Obtaining a gene expression matrix and corresponding spa- On the other hand, there are data that do not necessarily re-
tial coordinates from a raw ST data set is generally not a trivial pro- quire imaging but rather operate through capturing transcripts
cess and consists of a number of preprocessing steps. These steps within a defined spatial unit and linking them with a known coor-
are typically technology or platform dependent, but there are a dinate system before the sequencing step. As such, these approach-
few recurring preprocessing steps that are inherent to some or all es are typically less—or not—dependent on the raw image
technologies, such as image registration, stitching, and cell seg- processing steps described above. However additional steps are
mentation for data based on imaging (Fig. 2). needed after sequencing to map the transcript back to their spatial
For imaging-based ST data, such as the FISH and ISS technol- coordinates. When accompanying tissue images are available, they
ogies, the most frequent image processing steps are image correc- may be overlaid with the spatial coordinate system. The readers are
tion, stitching, registration, segmentation, followed by locating referred to the original protocols for more information about the
and decoding individual spots that usually correspond to a single data preprocessing procedures.
transcript. Initial corrections to the obtained images are almost al- Regardless of technological differences, a common goal in ST
ways needed to adjust for technological artifacts and are often de- analysis is to connect and integrate information from both gene
pendent on the experimental assay. The main goal here is to expression and cellular or transcript locations. This is crucial for
increase the signal-to-noise ratio and create normalized intensities extracting useful biological information, allowing linking with
for further downstream steps. Multiple overlapping fields of view, cell morphology and generating new hypotheses (Fig. 3). In the
or tiles, are needed when the tissue to be analyzed is too big in size, following sections we will review the state-of-the-art computation-
and they need to be stitched back together (Fig. 2A). Similarly, im- al methods and tools for these analyses. A curated list with addi-
ages that consist of multiple z-stacks can be misaligned because of tional details for all discussed methods is also provided at GitHub
technical or experimental procedures. For example, this occurs (https://github.com/drieslab/awesome-spatial-data-analysis).
when multiple hybridization rounds are sequentially imaged or
when creating a 3D data set by using adjacent 2D slices. The pro-
cess to correct for this misalignment is often referred to as image
Identification of cell types from ST data
registration and can be performed using a variety of different trans- Cell type identification and localization is probably the most basic
formation algorithms and strategies (Fig. 2B; Borovec et al. 2020). task for ST data analysis. If the data has single-cell resolution, such
As a demonstrating example, registration transformation is ap- as in multiplexed FISH approaches (Lubeck et al. 2014; Chen et al.
plied to a ST breast cancer data set (Andersson et al. 2020b), in 2015; Shah et al. 2016; Codeluppi et al. 2018; Moffitt et al. 2018;
which six sections were taken serially with a distance of 48 microns Eng et al. 2019; Kishi et al. 2019; Goh et al. 2020), unsupervised
from each other. Applying registration transformation on all sec- clustering combined with manual or automatic annotation is a
common approach to identify cell types in an unbiased manner
(Fig. 4A). Because the spatial information is not needed for cell
type identification, the task is highly similar to scRNA-seq analysis,
for which numerous methods have been developed (for a bench-
mark study, see Abdelaal et al. 2019). For example, community-
based methods such as Louvain (Blondel et al. 2008) and Leiden
clustering (Traag et al. 2019) are popular choices for cell type iden-
tification, in which the clustering results are used as initial guide
followed by often tedious manual biological annotations or
through automated workflows as recently discussed by Pasquini
Figure 1. Data sets used in this Perspective. et al. (2021). To show this approach, we used the MERFISH coronal

Genome Research 1707

www.genome.org
Dries et al.

A B C Apart from cell type annotation, meth-

ods have also been developed to impute
transcriptome-wide gene expression lev-
els by integration with scRNA-seq data
(Lopez et al. 2019; Lohoff et al. 2020).
Commercially available, array-
based ST technologies (such as 10x Ge-
nomics Visium and NanoString GeoMx)
typically do not have single-cell resolu-
tion. Because the variation of gene ex-
pression profiles may be associated with
changes of cell type composition rather
than new cell types, it is not appropriate
to apply a clustering algorithm directly to
such data and interpret the resulting
clusters as cell types. Furthermore, it is
possible to estimate cell type composi-
tion only if the underlying gene expres-
sion signatures are known. There are
two general approaches for estimating
cell type composition (Fig. 4C). The first
approach is to evaluate the enrichment
of cell type–specific markers among the
expressed genes at each spot (Moncada
Figure 2. Preprocessing of raw spatial transcriptomic data. (A) For spatial transcriptomics data paired et al. 2020; Dries et al. 2021). This ap-
with images, processing begins with correction and stitching of multiple captures or fields of views proach is fast and can be performed one
(FOVs) to form a clear composite image. (B) Images from multiple stacked sections of the same tissue
cell type at a time. However, the results
can be registered and the resulting spatial transformations mapped back to the transcriptomic data in
order to create an aligned 3D gene expression data set. This is illustrated with the breast cancer spatial are qualitative, indicating the presence
transcriptomics data set from Andersson et al. (2020b). (C) Several methods exist to provide expression or absence of a cell type. The second
data with spatial context. For technologies such as FISH and ISS that do not have clearly defined read approach, deconvolution, aims to quan-
spots or boundaries, cell segmentation (upper panel) is required in order to assign reads to individual cells. titatively estimate the proportion of
In situ capture or array-based methods, on the other hand (lower panel), assign reads to read spots based
on a spatial barcode unique to each spatial unit (e.g., spot). different cell types at each location.
Many deconvolution methods have
been developed and benchmarked for
slice data set and applied Leiden clustering, resulting in a total of RNA-seq data analysis (Avila Cobos et al. 2020). In principle, these
19 distinct clusters. These clusters are then annotated and mapped tools can also be applied to ST analysis. On the other hand, ST data
back to the spatial coordinates (Fig. 4B). have certain distinct properties; for example, the number of cells
Although it is possible to use a sequential FISH approach to associated with each location is often small. Therefore, it is often
generate transcriptome-scale profiles (Eng et al. 2019; Xia et al. more accurate to use methods that are tailored for ST analysis
2019), owing to the additional technical challenge, the common (Andersson et al. 2020a; Biancalani et al. 2020; Kleshchevnikov
practice is to target a limited number of genes (typically only a et al. 2020; Cable et al. 2021; Dong and Yuan 2021; Elosua-Bayes
few hundreds), which are often selected based on prior biological et al. 2021; Lopez et al. 2021; Song and Su 2021). Among these
knowledge. As a consequence, the data are insufficient to discover methods, RCTD uses a linear regression model for gene counts,
unknown cell types in an unbiased manner but allow the biolo- which further incorporates a random-effect term for platform-spe-
gists to annotate cell types whose gene signature is already known, cific variations (Cable et al. 2021). The gene expression levels are
often through external scRNA-seq analysis. Although the simplest modeled by a Poisson distribution. A similar approach is used in
approach is to identify the cell type whose gene signature has the stereoscope (Andersson et al. 2020a). Cell2location uses a similar
highest correlation, a drawback is that it does not distinguish cell approach but models gene expression using the negative binomial
type marker genes from the transcriptome-wide background. distribution (Kleshchevnikov et al. 2020). It can also model plat-
Numerous computational approaches have been developed to op- form- and location-specific effects. SpatialDWLS uses a two-step
timize accuracy. For example, one approach is to build a support procedure to reduce noise (Dong and Yuan 2021). The first step
vector machine classifier based on the scRNA-seq data but only identifies cell types that are likely to be present, by using an en-
use information from the subset of genes that is also profiled in richment analysis as described above, and then the second step
seqFISH (Zhu et al. 2018). A likelihood ratio test can also be used quantifies the relative proportion of each cell type by using a
(Vickovic et al. 2019). Importantly, cross-platform normalization dampened weighted least-square procedure previously developed
is needed to calibrate signals detected from different technologies. for RNA-seq data deconvolution (Tsoucas et al. 2019). SPOTlight
More generally, platform-specific technical variations can be esti- uses a seeded nonnegative matrix factorization (NMF) regression,
mated and reduced (Butler et al. 2018; Haghverdi et al. 2018; initialized using cell type marker genes and nonnegative least
Barkas et al. 2019; Hie et al. 2019; Korsunsky et al. 2019; Stuart squares (NNLS) for subsequent deconvolution (Elosua-Bayes
et al. 2019; Welch et al. 2019). Furthermore, Bayesian models et al. 2021). DSTG uses a graph-based convolutional network ap-
have been developed to incorporate the impact of cell segmenta- proach (Song and Su 2021). DestVI uses a variational inference ap-
tion uncertainty on cell type annotations (Qian et al. 2020). proach for deconvolution (Lopez et al. 2021). As an illustrating

1708 Genome Research

www.genome.org
 Spatial transcriptomic data analysis

A B C D that the total number of cells is known

(which can be extracted from the H&E
staining information), the deterministic
mode of Tangram alignment also serves
as a deconvolution method.

Characterizing spatial
E F G patterns of transcriptomic
profiles
The key contribution of ST analysis is to
characterize not just the cell types but
also how they are spatially organized.
This is fundamentally important for
studying the impact of tissue architecture
and cell–cell interactions (Fig. 5A,C,E).
Figure 3. Overview of spatial transcriptomics analysis methods. A variety of analyses can be performed
on spatial transcriptomics data. (A) Analysis can be performed on the image itself, ranging from early To study the spatial patterns associated
tasks such as cell segmentation to support of subcellular analysis through cell shape and size classification. with gene expression and cell states, pair-
(B) Cell types can be identified through clustering and annotation. Additional integration with external wise enrichment analysis can be used to
scRNA-seq data or deconvolution of spatial units that cover multiple cells (C) can be performed to fine-
identify cell type pairs that are likely to
tune cell type mapping. (D) The spatial distribution of cell types and the underlying cell-to-cell commu-
nication (E) can be computed. (F) Spatial expression patterns are identified and visualized based on in- be next to each other (Schapiro et al.
formation of gene expression and spatial coordinates. (G) Data at subcellular resolution can be used to 2017; Dries et al. 2021). Cell neighbor-
identify spatial and temporal dynamics of transcripts within a single cell. hood motif analysis identifies recurrent
patterns of multiple cell type neighbor-
hoods (Goltsev et al. 2018). An alterna-
example, we use the Visium heart data set and matching scRNA- tive approach to identify enriched patterns is to use topic models
seq data (Litviň uková et al. 2020) to perform both cell (Chen et al. 2020). Furthermore, the continuity of cell states can
type enrichment (Fig. 4D) and spatial deconvolution (Fig. 4E). Vi- be incorporated into a hidden Markov random field (HMRF) model
sualizing cell type enrichment is performed for each set of signa- to identify coherent spatial domains (Zhu et al. 2018). This ap-
ture genes, whereas deconvolution results in a quantitative proach has been extended in more recent studies (Chidester
assessment of cell type composition for each spot. et al. 2021; Zhao et al. 2021a). BayesSpace (Zhao et al. 2021a)
A complementary approach to study cell type localization is to uses a Bayesian formulation of HMRF, and the model parameters
use scRNA-seq data as the starting point and then reconstruct spa- are estimated by a Markov chain Monte Carlo (MCMC) algorithm,
tial information based on similarities with spatial expression pro- whereas SPICEMIX (Chidester et al. 2021) combines HMRF with
files. Before the explosion of ST technologies, it was possible to NMF. staNMF combines NMF with a stability criterion study to
obtain spatial information only for a handful of landmark genes identify spatial patterns (Wu et al. 2016). To illustrate how spatial
using traditional methods. Using such limited information, two network patterns and cellular neighborhoods are studied, we used
groups were able to reconstruct transcriptome-wide spatial patterns the MERFISH coronal slice data and created a cell–cell proximity
using clever computational modeling (Achim et al. 2015; Satija network based on the physical coordinates of each cell that are
et al. 2015). Around the same time, tomo-seq and Geo-seq technol- connected through Delaunay triangulation. The cell–cell proxim-
ogies were developed to reconstruct 3D patterns from gene expres- ity network along with the heatmap shows the closeness and con-
sion profiles obtained from 2D slices (Junker et al. 2014; Peng et al. nectivity between different cell types and informs users about the
2016). A key missing link is that the spatial information is not spatial topology of the studied tissue (Fig. 5B). A detailed explora-
directly measured from data; therefore, the patterns inferred from tion of individual niches is shown in Figure 5D. Here, specific cells
these analyses remain speculative. With the rapid development are identified as “source,” and then their connectivity with other
of ST technologies in the past few years, it is now possible to mea- neighboring cell types is depicted.
sure spatial information directly and further integrate with scRNA- A number of groups model spatial patterns of gene expression
seq data for additional refinement. Therefore, newer approaches in- as derived from predefined processes. For example, spatialDE uses a
tegrate scRNA-seq and ST data in a more balanced manner. For ex- random effect model that contains two terms, corresponding to
ample, a platform-agnostic, mutual nearest neighbor (MNN) the spatial and nonspatial component, respectively (Svensson
approach has been used to align these data types, which results et al. 2018a). The spatial component can be specified as various
in cell locations mapping (Haghverdi et al. 2018; Hie et al. 2019; forms such as linear, periodic, or a Gaussian process. The degree
Stuart et al. 2019). DEEPsc uses an artificial neural network to pre- of spatial variability is then quantified by the ratio of the variance
dict spatial locations (Maseda et al. 2021). GLUER combines joint explained by these two terms. SOMDE uses a similar approach but
NMF, MNN algorithm, and deep neural network to align data increases computational efficiency by first compressing spatial in-
(Peng et al. 2021). Tangram aligns scRNA-seq and ST data sets while formation by using a self-organizing map-based transformation
optimizing the spatial correlation between each gene in the scRNA- (Hao et al. 2021b). Trendsceek models spatial patterns as a marked
seq data and in the spatial data (Biancalani et al. 2020). A similar point process (Edsgärd et al. 2018). SPARK models spatial count
idea is also implemented in NovaSparc (Nitzan et al. 2019) and data through generalized linear spatial models with an additional
D-CE (Zhao et al. 2021b). Of note, the alignment can be either step to calibrate P-value calculation (Sun et al. 2020). Some meth-
probabilistic or deterministic. With the additional assumption ods are mainly concerned about local continuity. As an example,

Genome Research 1709

www.genome.org
Dries et al.

A B

Figure 4. Strategies for cell type identification with spatial transcriptomic data. (A) Spatial transcriptomics data at single-cell resolution can be directly
used to identify cell types in an analogous manner to scRNA-seq. In addition, external scRNA-seq from matching tissue can also be integrated to increase
the number of available features and aid in the identification of detected cell types. (B) An example of cell type annotation is shown on the MERFISH mouse
coronal brain slice data set. Each single dot represents a single cell, and colors indicate different cell types identified through clustering. A zoomed-in subset
shows the spatial cell type composition at a higher resolution. (C) Cell types in non-single-cell spatial transcriptomic data are identified through deconvo-
lution approaches that make use of external information or through gene enrichment strategies using sets of known marker genes or scRNA-seq informa-
tion. (D) Enrichment scores for two cell types within the human heart 10x Genomics Visium data set are overlaid on top of the spots within a region of
interest. (E) Pie charts depict the proportion of identified cell types within each selected spot used in D.

binSpect detects spatially coherent genes as those that tend to be based methods (Lubeck et al. 2014; Chen et al. 2015; Shah et al.
coexpressed in neighboring cells, using a spatial network formula- 2016; Codeluppi et al. 2018; Moffitt et al. 2018; Eng et al. 2019;
tion (Dries et al. 2021). Yet another approach is to quantify spatial Kishi et al. 2019; Goh et al. 2020), which are well known to have
structure in terms of diffusive steps it takes to reach a homoge- single-molecule resolution, ISS approaches (Lee et al. 2014; Wang
neous configuration (Anderson and Lundeberg 2021). The identi- et al. 2018; Qian et al. 2020; Hu et al. 2020b; Alon et al. 2021; Fu
fication of spatially coherent genes can in turn inform cell-state et al. 2021) also offer very high resolution. In addition, high-densi-
spatial pattern detection (Zhu et al. 2018). Alternatively, the spatial ty array or bead-based technologies (Vickovic et al. 2019; Alon et al.
gene and domain detection steps are inferred simultaneously (Hu 2021; Chen et al. 2021; Stickels et al. 2021) have also enabled sub-
et al. 2020a). As a concrete example, binSpect was used to identify cellular resolution. Here we use the seqFISH+ mouse somatosen-
genes with a spatial coherent pattern in the MERFISH coronal sory cortex data set to illustrate some key concepts of subcellular
brain slice data, and top-ranked genes are shown in Figure 5F. data analysis (Fig. 6). In a data set with subcellular resolution,
each point typically represents a single transcript (Fig. 6A). Analyz-
ing the subcellular gene expression patterns can be used as an alter-
Subcellular structure analysis native approach for spatial analysis but also can be used to enhance
With the advancement of newer technologies, it is now possible to the accuracy of cell segmentation (Fig. 6B). Finally, subcellular lo-
study subcellular transcript organizations. In addition to FISH- calization of RNA transcripts can also be used to gain biological

1710 Genome Research

www.genome.org
 Spatial transcriptomic data analysis

A B A number of methods have been de-

veloped to use subcellular gene ex-
pression patterns to circumvent cell
segmentation, which can be challenging.
For example, SSAM assigns cell type la-
bels directly to pixels without cell seg-
mentation (Park et al. 2021). stLearn
uses a similar approach but further clus-
ters spatially proximal pixels that are as-
signed to the same cell type (Pham et al.
2020). Spage2vec also uses a similar ap-
proach but adapts a neural network for-
mulation (Partel and Wählby 2021).
Alternatively, supervised cell type map-
ping strategies based on known cell
Spatial Plot with Cell–Cell Proximity Cell–Cell Proximity
Delaunay Network Network Scores type–specific signatures have been devel-
oped. For example, a naive Bayes model
C D is used to assign cell types for HDST
data (Vickovic et al. 2019). Subcellular
gene expression patterns can in turn be
used to improve cell segmentation. For
Other example, Baysor models the subcellular
Neighbor
Source gene expression patterns by using a Mar-
kov random field model and further inte-
grates cell shape labeling information
(such as DAPI) to improve cell segmenta-
tion accuracy (Petukhov et al. 2020).
Cell Neighborhood Sparcle (Prabhakaran et al. 2021) uses a
Analysis
E F Dirichlet process mixture model instead
as well as the transcripts’ distance be-
tween neighboring cells and adjacent
transcripts to enhance cell segmentation.
JTSA uses an EM algorithm to iteratively
improve pixel-level gene expression pro-
file classification and cell-boundary an-
notations (Littman et al. 2021).
Analysis of the subcellular patterns
of gene expression can also provide new
biological insights. For example, an in
situ RNA velocity approach has been
developed to use subcellular RNA locali-
zation information to infer the transcrip-
tion rates (Xia et al. 2019). Because newly
transcribed RNAs are cumulated in the
nucleus, whereas mature mRNA needs
to be transported to the cytoplasm for
translation (Fig. 6D), the relative compo-
sition of nuclear versus cytoplasmic tran-
Figure 5. Spatial pattern analyses. (A) Spatial distribution analysis of neighboring cell types. Network scripts associated with each gene can be
represents the likelihood of two cell types being found in close physical proximity to each other. (B) A used to estimate the transcriptional ac-
subset of cells from the MERFISH mouse coronal brain slice data set shows the spatial network connec-
tivity and cellular proximities between different cell types. (C ) At the single-cell level, cellular niches
tivity. This is performed by using a simi-
can be identified based on a target cell (yellow) and its direct neighboring cells (blue). The composition lar mathematical formulation as in the
and position of the neighboring cell types create a niche for the target cell (bottom). (D) Source and original RNA velocity paper (La Manno
neighboring cells are depicted within a small subset of the MERFISH mouse coronal brain slice data et al. 2018).
set. (E) Patterns based on spatial gene expression information are based on single or multiple genes
In addition, colocalized mRNA spe-
and are continuous (top) or discrete (bottom). (F) Individual genes with unique spatial coherent expres-
sion patterns in the MERFISH mouse brain coronal data set are shown on the right. cies in the cytoplasm can be identified
with high resolution by using direct
proximity labeling of RNA using the per-
insights that are not possible through cell-level analyses. Individual oxidase enzyme APEX2, a method called APEX-seq (Fazal et al.
spatial relationships between genes or between genes and subcellu- 2019). Analysis of the resulting data identifies a remarkable corre-
lar structures are found through analysis of colocalization patterns spondence between colocalized RNA with known protein colocal-
(Fig. 6C) and transcription dynamics within each cell (Fig. 6D). ization patterns (Fazal et al. 2019), suggesting RNA colocalization

Genome Research 1711

www.genome.org
Dries et al.

A B cell type compositions could lead to sig-

nificant changes of gene expression
even within the same cell type (Fig. 7B,C).
Giotto introduces a two-way compar-
ison method to identify interaction
changed genes by comparing the gene ex-
pression pattern between subsets of cells
within the same cell type but surrounded
by different neighboring cells (Dries et al.
C D 2021). Of note, using the spatial informa-
tion can significantly reduce the number
of false-positive ligand–receptor activity
predictions compared with using gene
expression information alone. A similar
approach is used in CellPhoneDB v3.0
(Garcia-Alonso et al. 2021). In this
study, the ST data do not have single-cell
resolution. To overcome this challenge,
the investigators applied Cell2location
Figure 6. Schematic diagram for spatial transcriptomics analysis at subcellular resolution. (A) For spa- (Kleshchevnikov et al. 2020) to infer the
tial data at subcellular resolution, each dot typically represents a single transcript or, alternatively, a spatial location of different cell types before
unit that is well below the cell size. (B) The location of each transcript, along with its gene identity, can be
used as input to try and segment each cell. (C) Individual transcripts can be colocalized with other tran- comparing gene expression patterns asso-
scripts (orange and blue) or with itself (green) or can be found at specific subcellular structures (pink at ciated with different cell neighborhoods.
membrane). (D) Transcription dynamics from individual or multiple genes can be inferred from the loca- Alternative approaches have been used to
tion of transcripts. Here nascent transcripts are typically found in the nucleus (blue), whereas processed quantify the effect of neighboring cell
transcripts are found in the cytoplasm (orange). The ratio between the two can provide an estimate for
the RNA velocity. Examples for each analysis are provided on the right of each panel using the seqFISH+
types, including convolutional neural
data set from the mouse somatosensory cortex. networks (Li et al. 2020; Yuan and Bar-Jo-
seph 2020), optimal transport (Cang and
Nie 2020), and multioutput regression
may facilitate local protein translation and complex formation (Fig. (Li et al. 2021). Another approach is to explicitly decompose a
6C). Also, mRNAs enriched in nuclear locations tend to code for gene expression profile into spatial and nonspatial components
proteins enriched in nuclear speckles and nucleoplasm. Alterna- and then use the cell type composition in the neighborhood to
tively, subcellular RNA colocalization can also be detected by AT- estimate the spatial components (Arnol et al. 2019). The analysis
LAS-seq, which uses sucrose density gradient ultracentrifugation of ligand–receptor interactions has also been extended to include
followed by RNA sequencing (Adekunle
and Wang 2020). In this study, it was
also found that RNAs tended to colocalize
A
with other RNAs in similar protein com-
plexes, in cellular compartments, or
with similar biological functions.

Understanding how cells

communicate with the tissue
environment
An important goal of ST analysis is to
study how cells communicate with the C B
tissue environment (Fig. 7). Cellular
behavior can be significantly affected by
the tissue environment through direct
physical interactions, secreted molecules,
or interactions with the extracellular
matrix (Fig. 7A). For example, the devel-
opment of tumor vasculature can signifi-
cantly promote tumor growth, whereas
enriched immune cells in tumor micro-
environments could significantly control
its proliferation. Cell–cell communica- Figure 7. Cellular communication inferred from ligand–receptor interactions. The known ligand–re-
ceptor interaction pairs are first explored using their gene expression profiles and then passed to a com-
tions are often spatially coordinated and
putational tool to generate communication scores that explain connectivity between and within each cell
can be highly cell type–specific (Armin- type as shown in A. A spatial graph can be constructed with these scores between different cell types as
gol et al. 2021). Thus, the variation of shown in B and C.

1712 Genome Research

www.genome.org
 Spatial transcriptomic data analysis

the effect of cofactors in the multiunit protein complexes to enhance images or sequence reads, because they are typically specific for
prediction accuracy (Jin et al. 2021). Of note, algorithms have also each technology, but we limit the survey to tools that are designed
been developed to reconstruct spatial locations from cell–cell inter- for downstream exploratory data analysis. Most of the code for
action patterns (Ren et al. 2020). these tools is written in the popular programming languages R
(R Core Team 2020) or Python or with a combination of both by
making use of recently developed interfaces such as reticulate
Integrative exploratory tools for spatial data analysis (https://github.com/rstudio/reticulate) or basilisk (http://basilisk
.fr), which allow developers to fully benefit from the strengths of
and visualization both worlds.
To effectively use and disseminate new methods that are being de- Giotto (Dries et al. 2021) is an R package that implements this
veloped to achieve a specific spatial data analysis task, it becomes latter strategy and has been shown to work on a large variety of ST
increasingly important to develop the necessary data structures technologies. It can also be applied to antibody-based protein mul-
and tools to work with them at a larger scale. Biologists will benefit tiplexed imaging technologies, although the latter is beyond the
from having integrative and interactive pipelines that allow them scope of this review. At its core, Giotto consists of an object specif-
to conduct various analysis steps, from importing raw data (Fig. ically designed for spatial data. At minimum, this object stores
8A) to image analysis (Fig. 8B), followed by the production of final both the count matrix and the accompanying 2D or 3D coordi-
analysis results and figures ready for publication (Fig. 8C), ideally nates of the spatial units, either individual cells or spatial aggre-
on their personal computer. Method developers can build on pre- gates as explained earlier. It provides routine analyses such as
vious spatial structures or make their new methods easily available filtering, clustering, and cell type annotation and presents spatial
to a larger audience. Currently, there are a number of comprehen- relationships as a network graph or through a spatial grid. This net-
sive toolboxes available, as described below. Here we will not dis- work forms the starting point for many new specific spatial analy-
cuss the specific steps necessary to process raw data, such as ses and facilitates the integration of other established algorithms

B Image Analysis

A B C
e e e
yp yp yp
Segmentation t l-t l-t
A
ll- l l
ce ce ce

Cellular Morphology and

Quantification

H&E image C Gene Exp.

p Analysis
y
Expression Data:

normalized
cell n
cell 2
cell 1

normalized
x y (z)
raw gene a gene b
gene 1 10 2 .... 0 cell 1 10 7 10 Cell type–specific markers
gene 2 5 7 10 cell 2 5 10 5 Clustering:
UMAP
UMAP2
....

....

tSNE
UMAP1

UMAP Clustering

gene n 2 10 2 cell n 7 5 10 Run Cell type

Enrichment
gene x cell cell
counts coordinates

Interactive
Spatial Grid
Interface
Integrative Exploratory Analysis Pipeline
Figure 8. An overview of interactive exploratory analysis pipeline. The integrative and interactive pipeline with several options can be used to analyze the
spatial data sets. (A) Spatial data analysis starts with importing and processing raw data sets. The analysis can then be subdivided into image-based analysis
(B) and gene expression–based analysis (C). Analysis based on images such as cell segmentation and morphological quantification is available to investigate
the cellular intricacies in a selected section of a tissue. Gene expression–based analysis consists of several approaches such as clustering, spatial network
construction, and cell type enrichment to visualize gene expression patterns. An interactive graphical interface makes these methods easier accessible
for novice users.

Genome Research 1713

www.genome.org
Dries et al.

through the creation of simple wrappers. For visualization purpos- Together, these tools play important roles for making the ST tech-
es, raw images of the profiled tissue can be stored and used to over- nologies broadly applicable.
lay the obtained spatial results. In parallel, Giotto offers a browser- Since the pioneering work by Ramón y Cajal, it has been stan-
based visualization tool, Giotto Viewer, that allows users to export dard practice to classify different cell types based on morphologi-
their obtained results and explore the spatial data set in an interac- cal changes. In recent years, there has been a paradigm shift by
tive manner. classifying cell types based on transcriptomic profiles, sometimes
Seurat is better known as a popular R package for scRNA-seq complemented by additional molecular modalities. Owing to the
analysis, but it commenced to offer some advanced functionalities rapid development of ST technologies, it is now possible to per-
through its spatial branch (Hao et al. 2021a). The functions are spe- form both transcriptomic profiling and morphology analyses for
cific to spatial data visualization and the identification of spatial the same cells, thereby providing a great opportunity to systemati-
expression patterns through the usage of established methods. cally investigate the relationship between these two fundamental-
Furthermore, other tools such as STUtility (Bergenstråhle et al. ly different approaches. A few methods have recently been
2020) and SPATA (Kueckelhaus et al. 2020a) have built on top of developed that focus on integration of both modalities (He et al.
the rich and performant data structure of Seurat to create more 2020; Tan et al. 2020; Gerbin et al. 2021). Although not directly re-
comprehensive pipelines that are currently specific for the ST tech- lated to spatial transcriptomics, an interesting finding from living
nology. STUtility (Bergenstråhle et al. 2020) was developed specif- imaging analysis indicates that changes in morphology might
ically for the ST technology and offers a wide variety of imaging even predict cell fate or state before this can be observed in the
and data analysis methods that are targeted for this approach. transcriptomic output (Buggenthin et al. 2017). Future work, in-
Similarly, SPATA (Kueckelhaus et al. 2020) focuses on ST data cluding the reconstruction of complete 3D tissues using CODA
and was developed to facilitate integration with the popular R (Kiemen et al. 2020), in this direction will help in bridging the
packages Seurat and Monocle. Besides visualization and common gap between communities.
data analysis functions, SPATA also has a rich repertoire of interac- An exciting new direction that is not covered here is spatial
tive methods to identify or delineate spatial trajectories. multiomics. New technology development has made it possible
Squidpy (Palla et al. 2021) is the spatial counterpart of to profile multiple modality information in the same cells while
SCANPY (Wolf et al. 2018), the popular Python library for preserving information, such as protein and RNA (Saka et al.
scRNA-seq analysis, and was created by the same laboratory. 2019; Liu et al. 2020; Merritt et al. 2020; Takei et al. 2021), intron
Similar to Giotto, it starts by representing the spatial information and mature mRNA (Shah et al. 2018; Mateo et al. 2019; Su et al.
through a spatial network and offers a large variety of downstream 2020), DNA, and RNA (Mondal et al. 2018; Mateo et al. 2019; Su
spatial analysis. In contrast to other toolboxes, it also provides et al. 2020; Takei et al. 2021). These technologies have made it pos-
analysis at the image level, which ranges from typical tasks such sible to analyze the correlation between different molecular modal-
as segmentation or registration to more advanced ways of extract- ities and offer mechanistic insights. Analyzing such data requires
ing and using morphology information in downstream analysis. development of novel computational methods and toolboxes. In
Stlearn (Pham et al. 2020) is another Python library for ST data fact, a number of multiomic analysis methods have already been
analysis with a specific focus on integrating both gene expression developed for sequencing-based assays (Argelaguet et al. 2018; But-
and image information through a joint representation. ler et al. 2018; Haghverdi et al. 2018; Barkas et al. 2019; Hie et al.
Most of these packages or toolboxes are developed in inde- 2019; Korsunsky et al. 2019; Stuart et al. 2019; Welch et al. 2019;
pendent laboratories, which results in multiple different data Biancalani et al. 2020; Peng et al. 2021). The readers are referred
structures that do not necessarily share the same data format. To to published reviews to learn more about this topic (Stuart and Sat-
overcome some of these challenges, the R/Bioconductor commu- ija 2019; Ma et al. 2020; Forcato et al. 2021). However, further devel-
nity is engaged in the careful design of generally applicable data opment is needed to incorporate the spatial context.
structures and has recently published the first version of the In sum, spatial technologies have brought many new chal-
spatialExperiment class (Righelli et al. 2021). This is a new S4 class lenges and opportunities. We believe that computational
that extends the popular singleCellExperiment class (Amezquita method development will continue to play a critical role in trans-
et al. 2020) and is designed to operate with several types of ST lating the promise of spatial technologies to reality by providing
data sets, including at both multi- and subcellular resolution. important tools for the analysis, visualization, and interpretation
Several spatial R packages already exist that use this data structure, of new data.
such as SpatialLIBD (Pardo et al. 2021) and Spaniel (Queen et al.
2019), which both excel in the creation of interactive R/Shiny
apps to visualize ST data sets. All together, these efforts could con-
tribute to the promotion of interoperability between these differ- Competing interest statement
ent toolboxes in the future.
The authors declare no competing interests.

Discussion
The rapid development of ST technologies has provided new op- Acknowledgments
portunities and challenges for data analysis. As summarized above, We thank Dr. Long Cai and his laboratory members (in particular,
there has been a lot of progress in this domain in recent years. Michal Polonsky, Arun Chakraborty, and Yujing Yang) for reading
Novel methods have been developed for attacking various ST-spe- our manuscript and helpful feedback. This research was supported
cific challenges. Integrative software packages have enabled biolo- by National Institutes of Health grants UH3HL145609
gists to easily analyze their own data from beginning to the end (Common Fund) and R01AG066028 (National Institute on
and to interactively explore the data via interactive visualization. Aging) (to G.-C.Y.).

1714 Genome Research

www.genome.org
 Spatial transcriptomic data analysis

References Butler A, Hoffman P, Smibert P, Papalexi E, Satija R. 2018. Integrating single-

cell transcriptomic data across different conditions, technologies, and
Abdelaal T, Michielsen L, Cats D, Hoogduin D, Mei H, Reinders MJT, species. Nat Biotechnol 36: 411–420. doi:10.1038/nbt.4096
Mahfouz A. 2019. A comparison of automatic cell identification meth- Cable DM, Murray E, Zou LS, Goeva A, Macosko EZ, Chen F, Irizarry RA.
ods for single-cell RNA sequencing data. Genome Biol 20: 194. doi:10 2021. Robust decomposition of cell type mixtures in spatial transcrip-
.1186/s13059-019-1795-z tomics. Nat Biotechnol doi:10.1038/s41587-021-00830-w
Achim K, Pettit J-B, Saraiva LR, Gavriouchkina D, Larsson T, Arendt D, Cang Z, Nie Q. 2020. Inferring spatial and signaling relationships between
Marioni JC. 2015. High-throughput spatial mapping of single-cell cells from single cell transcriptomic data. Nat Commun 11: 2084. doi:10
RNA-seq data to tissue of origin. Nat Biotechnol 33: 503–509. doi:10 .1038/s41467-020-15968-5
.1038/nbt.3209 Cao J, Spielmann M, Qiu X, Huang X, Ibrahim DM, Hill AJ, Zhang F,
Adekunle DA, Wang ET. 2020. Transcriptome-wide organization of subcel- Mundlos S, Christiansen L, Steemers FJ, et al. 2019. The single-cell tran-
lular microenvironments revealed by ATLAS-Seq. Nucleic Acids Res 48: scriptional landscape of mammalian organogenesis. Nature 566: 496–
5859–5872. doi:10.1093/nar/gkaa334 502. doi:10.1038/s41586-019-0969-x
Alon S, Goodwin DR, Sinha A, Wassie AT, Chen F, Daugharthy ER, Bando Y, Chen KH, Boettiger AN, Moffitt JR, Wang S, Zhuang X. 2015. RNA imaging:
Kajita A, Xue AG, Marrett K, et al. 2021. Expansion sequencing: spatially spatially resolved, highly multiplexed RNA profiling in single cells.
precise in situ transcriptomics in intact biological systems. Science 371: Science 348: aaa6090. doi:10.1126/science.aaa6090
eaax2656. doi:10.1126/science.aax2656 Chen Z, Soifer I, Hilton H, Keren L, Jojic V. 2020. Modeling multiplexed im-
Amezquita RA, Lun ATL, Becht E, Carey VJ, Carpp LN, Geistlinger L, Marini ages with Spatial-LDA reveals novel tissue microenvironments. J Comput
F, Rue-Albrecht K, Risso D, Soneson C, et al. 2020. Orchestrating single- Biol 27: 1204–1218. doi:10.1089/cmb.2019.0340
cell analysis with Bioconductor. Nat Methods 17: 137–145. doi:10.1038/ Chen A, Liao S, Ma K, Wu L, Lai Y, Yang J, Li W, Xu J. 2021. Large field of
s41592-019-0654-x view-spatially resolved transcriptomics at nanoscale resolution.
Anderson A, Lundeberg J. 2021. sepal: identifying transcript profiles with bioRxiv doi:10.1101/2021.01.17.427004
spatial patterns by diffusion-based modeling. Bioinformatics btab164. Chidester B, Zhou T, Ma J. 2021. SPICEMIX: integrative single-cell spatial
doi:10.1093/bioinformatics/btab164 modeling for inferring cell identity. bioRxiv doi:10.1101/2020.11.29
Andersson A, Bergenstråhle J, Asp M, Bergenstråhle L, Jurek A, Fernández .383067
Navarro J, Lundeberg J. 2020a. Single-cell and spatial transcriptomics Cho C-S, Xi J, Si Y, Park S-R, Hsu J-E, Kim M, Jun G, Kang HM, Lee JH. 2021.
enables probabilistic inference of cell type topography. Commun Biol Microscopic examination of spatial transcriptome using Seq-Scope. Cell
3: 565. doi:10.1038/s42003-020-01247-y 184: 3559–3572.e22. doi:10.1016/j.cell.2021.05.010
Andersson A, Larsson L, Stenbeck L, Salmén F, Ehinger A, Wu S, Al-Eryani G, Codeluppi S, Borm LE, Zeisel A, La Manno G, van Lunteren JA, Svensson CI,
Roden D, Swarbrick A, Borg Å, et al. 2020b. Spatial deconvolution of Linnarsson S. 2018. Spatial organization of the somatosensory cortex re-
HER2-positive breast tumors reveals novel intercellular relationships. vealed by osmFISH. Nat Methods 15: 932–935. doi:10.1038/s41592-018-
bioRxiv doi:10.1101/2020.07.14.200600 0175-z
Argelaguet R, Velten B, Arnol D, Dietrich S, Zenz T, Marioni JC, Buettner F, Dimopoulos S, Mayer CE, Rudolf F, Stelling J. 2014. Accurate cell segmenta-
Huber W, Stegle O. 2018. Multi-omics factor analysis: a framework for tion in microscopy images using membrane patterns. Bioinformatics 30:
unsupervised integration of multi-omics data sets. Mol Syst Biol 14: 2644–2651. doi:10.1093/bioinformatics/btu302
e8124. doi:10.15252/msb.20178124 Dong R, Yuan G-C. 2021. SpatialDWLS: accurate deconvolution of spatial
Armingol E, Officer A, Harismendy O, Lewis NE. 2021. Deciphering cell–cell transcriptomic data. Genome Biol 22: 145. doi:10.1186/s13059-021-
interactions and communication from gene expression. Nat Rev Genet 02362-7
22: 71–88. doi:10.1038/s41576-020-00292-x Dries R, Zhu Q, Dong R, Eng C-HL, Li H, Liu K, Fu Y, Zhao T, Sarkar A, Bao F,
Arnol D, Schapiro D, Bodenmiller B, Saez-Rodriguez J, Stegle O. 2019. et al. 2021. Giotto, a toolbox for integrative analysis and visualization of
Modeling cell–cell interactions from spatial molecular data with spatial spatial expression data. Genome Biol 22: 78. doi:10.1186/s13059-021-
variance component analysis. Cell Rep 29: 202–211.e6. doi:10.1016/j 02286-2
.celrep.2019.08.077 Edsgärd D, Johnsson P, Sandberg R. 2018. Identification of spatial expres-
Asp M, Bergenstråhle J, Lundeberg J. 2020. Spatially resolved transcrip- sion trends in single-cell gene expression data. Nat Methods 15: 339–
tomes-next generation tools for tissue exploration. Bioessays 42: 342. doi:10.1038/nmeth.4634
e1900221. doi:10.1002/bies.201900221 Elosua-Bayes M, Nieto P, Mereu E, Gut I, Heyn H. 2021. SPOTlight: seeded
Avila Cobos F, Alquicira-Hernandez J, Powell JE, Mestdagh P, De Preter K. NMF regression to deconvolute spatial transcriptomics spots with sin-
2020. Benchmarking of cell type deconvolution pipelines for transcrip- gle-cell transcriptomes. Nucleic Acids Res 49: e50. doi:10.1093/nar/
tomics data. Nat Commun 11: 5650. doi:10.1038/s41467-020-19015-1 gkab043
Barkas N, Petukhov V, Nikolaeva D, Lozinsky Y, Demharter S, Khodosevich Eng C-HL, Lawson M, Zhu Q, Dries R, Koulena N, Takei Y, Yun J, Cronin C,
K, Kharchenko PV. 2019. Joint analysis of heterogeneous single-cell Karp C, Yuan G-C, et al. 2019. Transcriptome-scale super-resolved imag-
RNA-seq dataset collections. Nat Methods 16: 695–698. doi:10.1038/ ing in tissues by RNA seqFISH+. Nature 568: 235–239. doi:10.1038/
s41592-019-0466-z s41586-019-1049-y
Berg S, Kutra D, Kroeger T, Straehle CN, Kausler BX, Haubold C, Schiegg M, Falk T, Mai D, Bensch R, Çiçek Ö, Abdulkadir A, Marrakchi Y, Böhm A,
Ales J, Beier T, Rudy M, et al. 2019. ilastik: interactive machine learning Deubner J, Jäckel Z, Seiwald K, et al. 2019. U-Net: deep learning for
for (bio)image analysis. Nat Methods 16: 1226–1232. doi:10.1038/ cell counting, detection, and morphometry. Nat Methods 16: 67–70.
s41592-019-0582-9 doi:10.1038/s41592-018-0261-2
Bergenstråhle J, Larsson L, Lundeberg J. 2020. Seamless integration of image Fazal FM, Han S, Parker KR, Kaewsapsak P, Xu J, Boettiger AN, Chang HY,
and molecular analysis for spatial transcriptomics workflows. BMC Ting AY. 2019. Atlas of subcellular RNA localization revealed by
Genomics 21: 482. doi:10.1186/s12864-020-06832-3 APEX-Seq. Cell 178: 473–490.e26. doi:10.1016/j.cell.2019.05.027
Biancalani T, Scalia G, Buffoni L, Avasthi R, Lu Z, Sanger A, Tokcan N, Forcato M, Romano O, Bicciato S. 2021. Computational methods for the in-
Vanderburg CR, Segerstolpe A, Zhang M, et al. 2020. Deep learning tegrative analysis of single-cell data. Brief Bioinform 22: 20–29. doi:10
and alignment of spatially-resolved whole transcriptomes of single cells .1093/bib/bbaa042
in the mouse brain with tangram. bioRxiv doi:10.1101/2020.08.29 Fu X, Sun L, Chen JY, Dong R, Lin Y, Palmiter RD, Lin S, Gu L. 2021.
.272831v3 Continuous polony gels for tissue mapping with high resolution and
Binnewies M, Roberts EW, Kersten K, Chan V, Fearon DF, Merad M, RNA capture efficiency. bioRxiv doi:10.1101/2021.03.17.435795
Coussens LM, Gabrilovich DI, Ostrand-Rosenberg S, Hedrick CC, et al. Garcia-Alonso L, Handfield LF, Roberts K. 2021. Mapping the temporal and
2018. Understanding the tumor immune microenvironment (TIME) spatial dynamics of the human endometrium in vivo and in vitro.
for effective therapy. Nat Med 24: 541–550. doi:10.1038/s41591-018- bioRxiv doi:10.1101/2021.01.02.425073
0014-x Gerbin KA, Grancharova T, Donovan-Maiye RM, Hendershott MC,
Blondel VD, Guillaume J-L, Lambiotte R, Lefebvre E. 2008. Fast unfolding of Anderson HG, Brown JM, Chen J, Dinh SQ, Gehring JL, Johnson GR,
communities in large networks. J Statist Mech Theor Exper 2008: P10008. et al. 2021. Cell states beyond transcriptomics: integrating structural or-
doi:10.1088/1742-5468/2008/10/P10008 ganization and gene expression in hiPSC-derived cardiomyocytes. Cell
Borovec J, Kybic J, Arganda-Carreras I, Sorokin DV, Bueno G, Khvostikov AV, Syst 12: 670–687.e10. doi:10.1016/j.cels.2021.05.001
Bakas S, Chang EI-C, Heldmann S, Kartasalo K, et al. 2020. ANHIR: au- Goh JJL, Chou N, Seow WY, Ha N, Cheng CPP, Chang Y-C, Zhao ZW, Chen
tomatic non-rigid histological image registration challenge. IEEE Trans KH. 2020. Highly specific multiplexed RNA imaging in tissues with
Med Imaging 39: 3042–3052. doi:10.1109/TMI.2020.2986331 split-FISH. Nat Methods 17: 689–693. doi:10.1038/s41592-020-0858-0
Buggenthin F, Buettner F, Hoppe PS, Endele M, Kroiss M, Strasser M, Goltsev Y, Samusik N, Kennedy-Darling J, Bhate S, Hale M, Vazquez G, Black
Schwarzfischer M, Loeffler D, Kokkaliaris KD, Hilsenbeck O, et al. S, Nolan GP. 2018. Deep profiling of mouse splenic architecture with
2017. Prospective identification of hematopoietic lineage choice by CODEX multiplexed imaging. Cell 174: 968–981.e15. doi:10.1016/j
deep learning. Nat Methods 14: 403–406. doi:10.1038/nmeth.4182 .cell.2018.07.010

Genome Research 1715

www.genome.org
Dries et al.

Greenwald NF, Miller G, Moen E, Kong A, Kagel A, Fullaway CC, McIntosh Li D, Ding J, Bar-Joseph Z. 2021. Identifying signaling genes in spatial sin-
BJ, Leow K, Schwartz MS, Dougherty T, et al. 2021. Whole-cell segmen- gle-cell expression data. Bioinformatics 37: 968–975. doi:10.1093/bioin
tation of tissue images with human-level performance using large-scale formatics/btaa769
data annotation and deep learning. bioRxiv doi:10.1101/2021.03.01 Liao J, Lu X, Shao X, Zhu L, Fan X. 2021. Uncovering an organ’s molecular
.431313v2 architecture at single-cell resolution by spatially resolved transcriptom-
Haghverdi L, Lun ATL, Morgan MD, Marioni JC. 2018. Batch effects in sin- ics. Trends Biotechnol 39: 43–58. doi:10.1016/j.tibtech.2020.05.006
gle-cell RNA-sequencing data are corrected by matching mutual nearest Littman R, Hemminger Z, Foreman R, Arneson D, Zhang G, Gómez-Pinilla
neighbors. Nat Biotechnol 36: 421–427. doi:10.1038/nbt.4091 F, Yang X, Wollman R. 2021. Joint cell segmentation and cell type anno-
Han X, Wang R, Zhou Y, Fei L, Sun H, Lai S, Saadatpour A, Zhou Z, Chen H, tation for spatial transcriptomics. Mol Syst Biol 17: e10108.
Ye F, et al. 2018. Mapping the mouse cell atlas by Microwell-seq. Cell doi:10.15252/msb.202010108
173: 1307. doi:10.1016/j.cell.2018.05.012 Litviň uková M, Talavera-López C, Maatz H, Reichart D, Worth CL, Lindberg
Han X, Zhou Z, Fei L, Sun H, Wang R, Chen Y, Chen H, Wang J, Tang H, Ge EL, Kanda M, Polanski K, Heinig M, Lee M, et al. 2020. Cells of the adult
W, et al. 2020. Construction of a human cell landscape at single-cell lev- human heart. Nature 588: 466–472. doi:10.1038/s41586-020-2797-4
el. Nature 581: 303–309. doi:10.1038/s41586-020-2157-4 Liu Y, Yang M, Deng Y, Su G, Enninful A, Guo CC, Tebaldi T, Zhang D, Kim
Hao Y, Hao S, Andersen-Nissen E, Mauck WM 3rd, Zheng S, Butler A, Lee MJ, D, Bai Z, et al. 2020. High-spatial-resolution multi-omics sequencing via
Wilk AJ, Darby C, Zagar M, et al. 2021a. Integrated analysis of multi- deterministic barcoding in tissue. Cell 183: 1665–1681.e18. doi:10
modal single-cell data. Cell 184: 3573–3587.e29. doi:10.1016/ .1016/j.cell.2020.10.026
j.cell.2021.04.048 Lohoff T, Ghazanfar S, Missarova A, Koulena N, Pierson N, Griffiths JA,
Hao M, Hua K, Zhang X. 2021b. SOMDE: a scalable method for identifying Bardot ES, Eng C-HL, Tyser RCV, Argelaguet R, et al. 2020. Highly mul-
spatially variable genes with self-organizing map. Bioinformatics tiplexed spatially resolved gene expression profiling of mouse organo-
btab471. doi:10.1093/bioinformatics/btab471 genesis. bioRxiv doi:10.1101/2020.11.20.391896
He B, Bergenstråhle L, Stenbeck L, Abid A, Andersson A, Borg Å, Maaskola J, Lopez R, Nazaret A, Langevin M, Samaran J, Regier J, Jordan MI, Yosef N.
Lundeberg J, Zou J. 2020. Integrating spatial gene expression and breast 2019. A joint model of unpaired data from scRNA-seq and spatial tran-
tumour morphology via deep learning. Nat Biomed Eng 4: 827–834. scriptomics for imputing missing gene expression measurements.
doi:10.1038/s41551-020-0578-x arXiv:1905.02269 [cs.LG].
Hie B, Bryson B, Berger B. 2019. Efficient integration of heterogeneous sin- Lopez R, Li B, Keren-Shaul H, Boyeau P, Kedmi M, Pilzer D, Jelinski A, David
gle-cell transcriptomes using Scanorama. Nat Biotechnol 37: 685–691. E, Wagner A, Addad Y, et al. 2021. Multi-resolution deconvolution of
doi:10.1038/s41587-019-0113-3 spatial transcriptomics data reveals continuous patterns of inflamma-
Hie B, Peters J, Nyquist SK, Shalek AK, Berger B, Bryson BD. 2020. tion. bioRxiv doi:10.1101/2021.05.10.443517
Computational methods for single-cell RNA sequencing. Annu Rev Lubeck E, Coskun AF, Zhiyentayev T, Ahmad M, Cai L. 2014. Single-cell in
Biomed Data Sci 3: 339–364. doi:10.1146/annurev-biodatasci-012220- situ RNA profiling by sequential hybridization. Nat Methods 11: 360–
100601 361. doi:10.1038/nmeth.2892
Hu J, Li X, Coleman K, Schroeder A, Irwin DJ, Lee EB, Shinohara RT, Li M. Ma A, McDermaid A, Xu J, Chang Y, Ma Q. 2020. Integrative methods and
2020a. Integrating gene expression, spatial location and histology to practical challenges for single-cell multi-omics. Trends Biotechnol 38:
identify spatial domains and spatially variable genes by graph convolu- 1007–1022. doi:10.1016/j.tibtech.2020.02.013
Marx V. 2021. Method of the year: spatially resolved transcriptomics. Nat
tional network. bioRxiv doi:10.1101/2020.11.30.405118
Methods 18: 9–14. doi:10.1038/s41592-020-01033-y
Hu KH, Eichorst JP, McGinnis CS, Patterson DM, Chow ED, Kersten K,
Maseda F, Cang Z, Nie Q. 2021. DEEPsc: a deep learning-based map connect-
Jameson SC, Gartner ZJ, Rao AA, Krummel MF. 2020b. ZipSeq: barcod-
ing single-cell transcriptomics and spatial imaging data. Front Genet 12:
ing for real-time mapping of single cell transcriptomes. Nat Methods
636743. doi:10.3389/fgene.2021.636743
17: 833–843. doi:10.1038/s41592-020-0880-2
Mateo LJ, Murphy SE, Hafner A, Cinquini IS, Walker CA, Boettiger AN.
Jin S, Guerrero-Juarez CF, Zhang L, Chang I, Ramos R, Kuan C-H, Myung P,
2019. Visualizing DNA folding and RNA in embryos at single-cell reso-
Plikus MV, Nie Q. 2021. Inference and analysis of cell–cell communica-
lution. Nature 568: 49–54. doi:10.1038/s41586-019-1035-4
tion using cellChat. Nat Commun 12: 1088. doi:10.1038/s41467-021-
Merritt CR, Ong GT, Church SE, Barker K, Danaher P, Geiss G, Hoang M,
21246-9
Jung J, Liang Y, McKay-Fleisch J, et al. 2020. Multiplex digital spatial pro-
Junker JP, Noël ES, Guryev V, Peterson KA, Shah G, Huisken J, McMahon AP,
filing of proteins and RNA in fixed tissue. Nat Biotechnol 38: 586–599.
Berezikov E, Bakkers J, van Oudenaarden A. 2014. Genome-wide RNA doi:10.1038/s41587-020-0472-9
tomography in the zebrafish embryo. Cell 159: 662–675. doi:10.1016/ Moffitt JR, Bambah-Mukku D, Eichhorn SW, Vaughn E, Shekhar K, Perez JD,
j.cell.2014.09.038 Rubinstein ND, Hao J, Regev A, Dulac C, et al. 2018. Molecular, spatial,
Kiemen A, Braxton AM, Grahn MP, Han KS, Babu JM, Reichel R, Amoa F, and functional single-cell profiling of the hypothalamic preoptic re-
Hong S-M, Cornish TC, Thompson ED, et al. 2020. In situ characteriza- gion. Science 362: eaau5324. doi:10.1126/science.aau5324
tion of the 3D microanatomy of the pancreas and pancreatic cancer at Moncada R, Barkley D, Wagner F, Chiodin M, Devlin JC, Baron M, Hajdu
single cell resolution. bioRxiv doi:10.1101/2020.12.08.416909 CH, Simeone DM, Yanai I. 2020. Integrating microarray-based spatial
Kishi JY, Lapan SW, Beliveau BJ, West ER, Zhu A, Sasaki HM, Saka SK, Wang transcriptomics and single-cell RNA-seq reveals tissue architecture in
Y, Cepko CL, Yin P. 2019. SABER amplifies FISH: enhanced multiplexed pancreatic ductal adenocarcinomas. Nat Biotechnol 38: 333–342.
imaging of RNA and DNA in cells and tissues. Nat Methods 16: 533–544. doi:10.1038/s41587-019-0392-8
doi:10.1038/s41592-019-0404-0 Mondal M, Liao R, Nazaroff CD, Samuel AD, Guo J. 2018. Highly multi-
Kleshchevnikov V, Shmatko A, Dann E, Aivazidis A, King HW, Li T, Lomakin plexed single-cell in situ RNA and DNA analysis with bioorthogonal
A, Kedlian V, Jain MS, Park JS, et al. 2020. Comprehensive mapping of cleavable fluorescent oligonucleotides. Chem Sci 9: 2909–2917. doi:10
tissue cell architecture via integrated single cell and spatial transcrip- .1039/C7SC05089E
tomics. bioRxiv doi:10.1101/2020.11.15.378125 Nitzan M, Karaiskos N, Friedman N, Rajewsky N. 2019. Gene expression car-
Korsunsky I, Millard N, Fan J, Slowikowski K, Zhang F, Wei K, Baglaenko Y, tography. Nature 576: 132–137. doi:10.1038/s41586-019-1773-3
Brenner M, Loh P-R, Raychaudhuri S. 2019. Fast, sensitive and accurate Packer JS, Zhu Q, Huynh C, Sivaramakrishnan P, Preston E, Dueck H,
integration of single-cell data with harmony. Nat Methods 16: 1289– Stefanik D, Tan K, Trapnell C, Kim J, et al. 2019. A lineage-resolved mo-
1296. doi:10.1038/s41592-019-0619-0 lecular atlas of C. elegans embryogenesis at single-cell resolution. Science
Kueckelhaus J, von Ehr J, Ravi VM, Will P, Joseph K, Beck J, Hofmann UG, 365: eaax1971. doi:10.1126/science.aax1971
Delev D, Schnell O, Heiland DH. 2020. Inferring spatially transient Palla G, Spitzer H, Klein M, Fischer D, Schaar AC, Kuemmerle LB, Rybakov S,
gene expression pattern from spatial transcriptomic studies. bioRxiv Ibarra IL, Holmberg O, Virshup I, et al. 2021. Squidpy: a scalable frame-
doi:10.1101/2020.10.20.346544 work for spatial single cell analysis. bioRxiv doi:10.1101/2021.02.19
La Manno G, Soldatov R, Zeisel A, Braun E, Hochgerner H, Petukhov V, .431994
Lidschreiber K, Kastriti ME, Lönnerberg P, Furlan A, et al. 2018. RNA ve- Pardo B, Spangler A, Weber LM, Hicks SC, Jaffe AE, Martinowich K, Maynard
locity of single cells. Nature 560: 494–498. doi:10.1038/s41586-018- KR, Collado-Torres L. 2021. spatialLIBD: an R/Bioconductor package to
0414-6 visualize spatially-resolved transcriptomics data. bioRxiv doi:10.1101/
Lee JH, Daugharthy ER, Scheiman J, Kalhor R, Yang JL, Ferrante TC, Terry R, 2021.04.29.440149
Jeanty SSF, Li C, Amamoto R, et al. 2014. Highly multiplexed subcellular Park J, Choi W, Tiesmeyer S, Long B, Borm LE, Garren E, Nguyen TN, Tasic B,
RNA sequencing in situ. Science 343: 1360–1363. doi:10.1126/science Codeluppi S, Graf T, et al. 2021. Cell segmentation-free inference of cell
.1250212 types from in situ transcriptomics data. Nat Commun 12: 3545.
Li D, Ma Q, Chen J, Liu A, Cheung J, Xie Y, Gudjonson H, Nawy T, Pe’er D, doi:10.1038/s41467-021-23807-4
Pe’er I. 2020. Counterfactual hypothesis testing of tumor microenviron- Partel G, Wählby C. 2021. Spage2vec: unsupervised representation of local-
ment scenarios through semantic image synthesis. bioRxiv doi:10 ized spatial gene expression signatures. FEBS J 288: 1859–1870. doi:10
.1101/2020.10.27.358101 .1111/febs.15572

1716 Genome Research

www.genome.org
 Spatial transcriptomic data analysis

Pasquini G, Rojo Arias JE, Schäfer P, Busskamp V. 2021. Automated methods Spanjaard B, Hu B, Mitic N, Olivares-Chauvet P, Janjuha S, Ninov N, Junker
for cell type annotation on scRNA-seq data. Comput Struct Biotechnol J JP. 2018. Simultaneous lineage tracing and cell-type identification using
19: 961–969. doi:10.1016/j.csbj.2021.01.015 CRISPR-Cas9-induced genetic scars. Nat Biotechnol 36: 469–473. doi:10
Peng G, Suo S, Chen J, Chen W, Liu C, Yu F, Wang R, Chen S, Sun N, Cui G, .1038/nbt.4124
et al. 2016. Spatial transcriptome for the molecular annotation of line- Ståhl PL, Salmén F, Vickovic S, Lundmark A, Navarro JF, Magnusson J,
age fates and cell identity in mid-gastrula mouse embryo. Dev Cell 36: Giacomello S, Asp M, Westholm JO, Huss M, et al. 2016. Visualization
681–697. doi:10.1016/j.devcel.2016.02.020 and analysis of gene expression in tissue sections by spatial transcrip-
Peng T, Chen G, Tan K. 2021. GLUER: integrative analysis of single-cell tomics. Science 353: 78–82. doi:10.1126/science.aaf2403
omics and imaging data by deep neural network. bioRxiv doi:10 Stickels RR, Murray E, Kumar P, Li J, Marshall JL, Di Bella DJ, Arlotta P,
.1101/2021.01.25.427845 Macosko EZ, Chen F. 2021. Highly sensitive spatial transcriptomics at
Perkel JM. 2019. Starfish enterprise: finding RNA patterns in single cells. near-cellular resolution with slide-seqV2. Nat Biotechnol 39: 313–319.
Nature 572: 549–551. doi:10.1038/d41586-019-02477-9 doi:10.1038/s41587-020-0739-1
Petukhov V, Soldatov RA, Khodosevich K, Kharchenko PV. 2020. Bayesian Stringer C, Wang T, Michaelos M, Pachitariu M. 2021. Cellpose: a generalist
segmentation of spatially resolved transcriptomics data. bioRxiv algorithm for cellular segmentation. Nat Methods 18: 100–106. doi:10
doi:10.1101/2020.10.05.326777 .1038/s41592-020-01018-x
Pham D, Tan X, Xu J, Grice LF, Lam PY, Raghubar A, Vukovic J, Ruitenberg Stuart T, Satija R. 2019. Integrative single-cell analysis. Nat Rev Genet 20:
MJ, Nguyen Q. 2020. stLearn: integrating spatial location, tissue mor- 257–272. doi:10.1038/s41576-019-0093-7
phology and gene expression to find cell types, cell–cell interactions Stuart T, Butler A, Hoffman P, Hafemeister C, Papalexi E, Mauck WM 3rd,
and spatial trajectories within undissociated tissues. bioRxiv doi:10 Hao Y, Stoeckius M, Smibert P, Satija R. 2019. Comprehensive integra-
.1101/2020.05.31.125658 tion of single-cell data. Cell 177: 1888–1902.e21. doi:10.1016/j.cell
Pijuan-Sala B, Griffiths JA, Guibentif C, Hiscock TW, Jawaid W, Calero-Nieto .2019.05.031
FJ, Mulas C, Ibarra-Soria X, Tyser RCV, Ho DLL, et al. 2019. A single-cell Su J-H, Zheng P, Kinrot SS, Bintu B, Zhuang X. 2020. Genome-scale imaging
molecular map of mouse gastrulation and early organogenesis. Nature of the 3D organization and transcriptional activity of chromatin. Cell
566: 490–495. doi:10.1038/s41586-019-0933-9 182: 1641–1659.e26. doi:10.1016/j.cell.2020.07.032
Prabhakaran S, Nawy T, Pe’er’ D. 2021. Sparcle: assigning tran- Sun S, Zhu J, Zhou X. 2020. Statistical analysis of spatial expression patterns
scripts to cells in multiplexed images. bioRxiv doi:10.1101/2021.02.13 for spatially resolved transcriptomic studies. Nat Methods 17: 193–200.
.431099 doi:10.1038/s41592-019-0701-7
Qian X, Harris KD, Hauling T, Nicoloutsopoulos D, Muñoz-Manchado AB, Svensson V, Teichmann SA, Stegle O. 2018a. SpatialDE: identification of
Skene N, Hjerling-Leffler J, Nilsson M. 2020. Probabilistic cell typing en- spatially variable genes. Nat Methods 15: 343–346. doi:10.1038/nmeth
ables fine mapping of closely related cell types in situ. Nat Methods 17: .4636
101–106. doi:10.1038/s41592-019-0631-4 Svensson V, Vento-Tormo R, Teichmann SA. 2018b. Exponential scaling of
Queen R, Cheung K, Lisgo S, Coxhead J, Cockell S. 2019. Spaniel: analysis single-cell RNA-seq in the past decade. Nat Protoc 13: 599–604. doi:10
and interactive sharing of spatial transcriptomics data. bioRxiv doi:10 .1038/nprot.2017.149
.1101/619197 Tabula Muris Consortium, Overall coordination, Logistical coordination,
R Core Team. 2020. R: a language and environment for statistical computing. R Organ collection and processing, Library preparation and sequencing,
Foundation for Statistical Computing, Vienna. https://www.R-project Computational data analysis, Cell type annotation, Writing group,
.org/. Supplemental text writing group, Principal investigators. 2018. Single-
Ren X, Zhong G, Zhang Q, Zhang L, Sun Y, Zhang Z. 2020. Reconstruction cell transcriptomics of 20 mouse organs creates a Tabula Muris. Nature
of cell spatial organization from single-cell RNA sequencing data based 562: 367–372. doi:10.1038/s41586-018-0590-4
on ligand-receptor mediated self-assembly. Cell Res 30: 763–778. doi:10 Takei Y, Yun J, Zheng S, Ollikainen N, Pierson N, White J, Shah S, Thomassie
.1038/s41422-020-0353-2 J, Suo S, Eng C-HL, et al. 2021. Integrated spatial genomics reveals global
Righelli D, Weber LM, Crowell HL, Pardo B, Collado-Torres L, Ghazanfar S, architecture of single nuclei. Nature 590: 344–350. doi:10.1038/s41586-
Lun ATL, Hicks SC, Risso D. 2021. SpatialExperiment: infrastructure for 020-03126-2
spatially resolved transcriptomics data in R using Bioconductor. bioRxiv Tan X, Su A, Tran M, Nguyen Q. 2020. SpaCell: integrating tissue morphol-
doi:10.1101/2021.01.27.428431v2 ogy and spatial gene expression to predict disease cells. Bioinformatics
Rodriques SG, Stickels RR, Goeva A, Martin CA, Murray E, Vanderburg CR, 36: 2293–2294. doi:10.1093/bioinformatics/btz914
Welch J, Chen LM, Chen F, Macosko EZ. 2019. Slide-seq: a scalable tech- Traag VA, Waltman L, van Eck NJ. 2019. From Louvain to Leiden: guarantee-
nology for measuring genome-wide expression at high spatial resolu- ing well-connected communities. Sci Rep 9: 5233. doi:10.1038/s41598-
tion. Science 363: 1463–1467. doi:10.1126/science.aaw1219 019-41695-z
Saka SK, Wang Y, Kishi JY, Zhu A, Zeng Y, Xie W, Kirli K, Yapp C, Cicconet Tsoucas D, Dong R, Chen H, Zhu Q, Guo G, Yuan G-C. 2019. Accurate esti-
M, Beliveau BJ, et al. 2019. Immuno-SABER enables highly multiplexed mation of cell-type composition from gene expression data. Nat
and amplified protein imaging in tissues. Nat Biotechnol 37: 1080–1090. Commun 10: 2975. doi:10.1038/s41467-019-10802-z
doi:10.1038/s41587-019-0207-y Vicar T, Balvan J, Jaros J, Jug F, Kolar R, Masarik M, Gumulec J. 2019. Cell
Satija R, Farrell JA, Gennert D, Schier AF, Regev A. 2015. Spatial reconstruc- segmentation methods for label-free contrast microscopy: review and
tion of single-cell gene expression data. Nat Biotechnol 33: 495–502. comprehensive comparison. BMC Bioinformatics 20: 360. doi:10.1186/
doi:10.1038/nbt.3192 s12859-019-2880-8
Schapiro D, Jackson HW, Raghuraman S, Fischer JR, Zanotelli VRT, Schulz Vickovic S, Eraslan G, Salmén F, Klughammer J, Stenbeck L, Schapiro D, Äijö
D, Giesen C, Catena R, Varga Z, Bodenmiller B. 2017. histoCAT: analysis T, Bonneau R, Bergenstråhle L, Navarro JF, et al. 2019. High-definition
of cell phenotypes and interactions in multiplex image cytometry data. spatial transcriptomics for in situ tissue profiling. Nat Methods 16:
Nat Methods 14: 873–876. doi:10.1038/nmeth.4391 987–990. doi:10.1038/s41592-019-0548-y
Schmidt U, Weigert M, Broaddus C, Myers G. 2018. Cell detection with star- Wang X, Allen WE, Wright MA, Sylwestrak EL, Samusik N, Vesuna S, Evans
convex polygons. In Medical Image Computing and Computer Assisted K, Liu C, Ramakrishnan C, Liu J, et al. 2018. Three-dimensional intact-
Intervention – MICCAI 2018. Lecture Notes in Computer Science (ed. tissue sequencing of single-cell transcriptional states. Science 361:
Frangi A, et al.), Vol. 11071. Springer, Cham. doi:10.1007/978-3-030- eaat5691. doi:10.1126/science.aat5691
00934-2_30 Welch JD, Kozareva V, Ferreira A, Vanderburg C, Martin C, Macosko EZ.
Sebé-Pedrós A, Saudemont B, Chomsky E, Plessier F, Mailhé M-P, Renno 2019. Single-cell multi-omic integration compares and contrasts fea-
J, Loe-Mie Y, Lifshitz A, Mukamel Z, Schmutz S, et al. 2018. Cnidarian tures of brain cell identity. Cell 177: 1873–1887.e17. doi:10.1016/j
cell type diversity and regulation revealed by whole-organism .cell.2019.05.006
single-cell RNA-Seq. Cell 173: 1520–1534.e20. doi:10.1016/j.cell.2018 Whiteside TL. 2008. The tumor microenvironment and its role in
.05.019 promoting tumor growth. Oncogene 27: 5904–5912. doi:10.1038/onc
Shah S, Lubeck E, Zhou W, Cai L. 2016. In situ transcription profiling of sin- .2008.271
gle cells reveals spatial organization of cells in the mouse hippocampus. Wolf FA, Angerer P, Theis FJ. 2018. SCANPY: large-scale single-cell gene ex-
Neuron 92: 342–357. doi:10.1016/j.neuron.2016.10.001 pression data analysis. Genome Biol 19: 15. doi:10.1186/s13059-017-
Shah S, Takei Y, Zhou W, Lubeck E, Yun J, Eng C-HL, Koulena N, Cronin C, 1382-0
Karp C, Liaw EJ, et al. 2018. Dynamics and spatial genomics of the na- Wu S, Joseph A, Hammonds AS, Celniker SE, Yu B, Frise E. 2016. Stability-
scent transcriptome by intron seqFISH. Cell 174: 363–376.e16. doi:10 driven nonnegative matrix factorization to interpret spatial gene expres-
.1016/j.cell.2018.05.035 sion and build local gene networks. Proc Natl Acad Sci 113: 4290–4295.
Song Q, Su J. 2021. DSTG: deconvoluting spatial transcriptomics data doi:10.1073/pnas.1521171113
through graph-based artificial intelligence. Brief Bioinform bbaa414. Xia C, Fan J, Emanuel G, Hao J, Zhuang X. 2019. Spatial transcriptome pro-
doi:10.1093/bib/bbaa414 filing by MERFISH reveals subcellular RNA compartmentalization and

Genome Research 1717

www.genome.org
Dries et al.

cell cycle-dependent gene expression. Proc Natl Acad Sci 116: 19490– at subspot resolution with BayesSpace. Nat Biotechnol doi:10.1038/
19499. doi:10.1073/pnas.1912459116 s41587-021-00935-2
Yuan G-C. 2019. Computational methods for single-cell data analysis. Humana Zhao Y, Zhang S, Cannistraci CV, Han J-DJ. 2021b. De novo spatial reconstruc-
Press, New York. tion of single cells by developmental coalescent embedding of transcrip-
Yuan Y, Bar-Joseph Z. 2020. GCNG: graph convolutional networks for infer- tomic networks. Preprints doi:10.20944/preprints202103.0196.v1
ring gene interaction from spatial transcriptomics data. Genome Biol 21: Zhu Q, Shah S, Dries R, Cai L, Yuan G-C. 2018. Identification of spatially as-
300. doi:10.1186/s13059-020-02214-w sociated subpopulations by combining scRNAseq and sequential fluo-
Zhao E, Stone MR, Ren X, Guenthoer J, Smythe KS, Pulliam T, Williams SR, rescence in situ hybridization data. Nat Biotechnol 36: 1183–1190.
Uytingco CR, Taylor SEB, Nghiem P, et al. 2021a. Spatial transcriptomics doi:10.1038/nbt.4260

1718 Genome Research

www.genome.org