Skip to main content
NCBI home page
Bioinformatics logoLink to Bioinformatics
. 2023 Sep 12;39(9):btad550. doi: 10.1093/bioinformatics/btad550

MoleculeExperiment enables consistent infrastructure for molecule-resolved spatial omics data in bioconductor

Bárbara Zita Peters Couto 1,2,3, Nicholas Robertson 4,5,6,7, Ellis Patrick 8,9,10,11,12,✉, Shila Ghazanfar 13,14,15,✉
Editor: Anthony Mathelier
PMCID: PMC10504467  PMID: 37698995

Abstract

Motivation

Imaging-based spatial transcriptomics (ST) technologies have achieved subcellular resolution, enabling detection of individual molecules in their native tissue context. Data associated with these technologies promise unprecedented opportunity toward understanding cellular and subcellular biology. However, in R/Bioconductor, there is a scarcity of existing computational infrastructure to represent such data, and particularly to summarize and transform it for existing widely adopted computational tools in single-cell transcriptomics analysis, including SingleCellExperiment and SpatialExperiment (SPE) classes. With the emergence of several commercial offerings of imaging-based ST, there is a pressing need to develop consistent data structure standards for these technologies at the individual molecule-level.

Results

To this end, we have developed MoleculeExperiment, an R/Bioconductor package, which (i) stores molecule and cell segmentation boundary information at the molecule-level, (ii) standardizes this molecule-level information across different imaging-based ST technologies, including 10× Genomics’ Xenium, and (iii) streamlines transition from a MoleculeExperiment object to a SpatialExperiment object. Overall, MoleculeExperiment is generally applicable as a data infrastructure class for consistent analysis of molecule-resolved spatial omics data.

Availability and implementation

The MoleculeExperiment package is publicly available on Bioconductor at https://bioconductor.org/packages/release/bioc/html/MoleculeExperiment.html. Source code is available on Github at: https://github.com/SydneyBioX/MoleculeExperiment. The vignette for MoleculeExperiment can be found at https://bioconductor.org/packages/release/bioc/html/MoleculeExperiment.html.

1 Introduction

Spatial omics is a maturing field, especially imaging-based spatial transcriptomics (ST) technologies (Moffitt et al. 2022; Wu et al. 2022). Since the publication of single-molecule FISH, many imaging-based ST technologies have been developed (Williams et al. 2022). While the transcriptome coverage of these technologies is not complete, they enable cellular and even subcellular resolution (Wu et al. 2022). In addition, various imaging-based ST technologies have recently started to be commercially shipped, such as 10× Genomics’s Xenium (Janesick et al. 2022), NanoString CosMX, and Vizgen MERSCOPE, and thus their use is expected to massively increase in scale. Imaging-based ST has been employed in multiple studies, including investigating the progression from ductal carcinoma in situ to invasive carcinoma (Janesick et al. 2022), analyzing the complex immune landscape of the tumor microenvironment in lung tumors (Chen et al. 2023), and facilitating the creation of the first comprehensive spatial atlas of the mouse brain (Yao et al. 2023). The highly informative cellular and subcellular resolution of imaging-based ST, as well as its increasing commercial availability, motivate the generation of software that helps scientists consistently handle this type of high resolution data.

Recently, SpatialExperiment (SPE) was developed as an object class for the study of ST data (Righelli et al. 2022). Just like the commonly used SingleCellExperiment class, SPE aims to promote reproducibility of analyses and interoperability of different software on the same data (Amezquita et al. 2020; Righelli et al. 2022). Moreover, by being a part of the Bioconductor project, these packages are a part of an effort to disseminate open data analysis and promote software maintenance and enhancement in the life sciences (Amezquita et al. 2020). However, the SPE package, and more recent extension SpatialFeatureExperiment (Moses et al. 2023), only allows the storage of gene expression information at the cell or spot level. For molecules that are assigned to cells, their specific locations can be kept via BumpyMatrix assay (https://bioconductor.org/packages/release/bioc/html/BumpyMatrix.html). However, molecules that have not been assigned to a cell, between 5% and 30% of all molecules in a given tissue (Supplementary Table 1), by the cell segmentation method are lost, which is disadvantageous, as these transcripts could yield valuable biological insights (Prabhakaran 2022). For example, performing a region-level differential expression analysis could be more accurate if all detected transcripts are taken into account, even in spaces where transcripts have not been assigned to a cell. However, if one were interested in doing such an analysis on ST data that has been summarized as an SPE object, one would only be able to do this at the cell-level. Therefore, to leverage the molecule resolution of recent technologies, there is a need for a class that avoids premature summarization of ST data, and enables analysis of transcripts in their spatial locations irrespective of cellular compartmentalization.

Beyond the Bioconductor project, there are varied efforts to represent molecule-resolved ST data. The Python package Squidpy (Palla et al. 2022) does not handle ST data at the molecule-level, but instead assumes presence of a spatial cell-by-gene counts matrix, while the Seurat R package (Hao et al. 2023) has recently been extended to contain molecules as a slot containing an “sp” class object for visualization alongside existing cell-by-gene expression matrices. There have been very recent efforts to build upon image data file formats to include molecule and segmentation information via the OME-NGFF Project (Moore et al. 2023), alongside Python package SpatialData (Marconato et al. 2023) to read and operate with these data. Overall, there is a collective shift toward representing molecule-resolved ST data beyond the cell-by-gene level, but there is a lack of representation of this currently in R, and particularly in the Bioconductor Project.

In this article, we introduce the MoleculeExperiment class, which represents ST data at the molecule-level. In addition, the MoleculeExperiment class imposes standardized data formats and terminology to avoid the need for manual file conversion and complex analysis scripts of molecule-based ST data. Moreover, the MoleculeExperiment package facilitates the transition to a cell-level analysis with the already existing SpatialExperiment class. Here, we enable the application of MoleculeExperiment to Xenium (10× Genomics), CosMx (NanoString), and MERSCOPE (Vizgen) data. In summary, the MoleculeExperiment package aims to facilitate the downstream analysis of different imaging-based ST data, both at the molecule-level and cellular-level, with the large diversity of data analysis tools in the Bioconductor project.

2 Methods

2.1 Examination of vendors’ public molecule-resolved spatial transcriptomics data bundles

We examined molecule and boundary data structures from the following technologies: 10× Genomics Xenium, NanoString CosMx, and Vizgen MERSCOPE. We used these vendors’ publicly available output data bundles, in some cases requiring a minimal sign in or form completion. We used these data bundles to inform our readXenium, readCosmx, and readMerscope functions, respectively. In particular, Xenium data correspond to three replicates from fresh frozen mouse brain tissue, accessed online on 8 February 2023; CosMx data correspond to human non-small cell lung cancer, accessed on 27 February 2023; and MERSCOPE data are from human ovarian cancer, accessed on 27 February 2023.

We assessed commonalities in terms of the detected transcripts files as well as cell boundary or segmentation files. No commonalities were found in the cell boundary files across the technologies. Vizgen’s output bundle contains several hdf5 files, Xenium a single csv.gz file, and NanoString has no single file with cell boundaries, but shares the identified cell IDs between the transcript, count matrix, and cell metadata files instead.

2.2 Assessing memory requirements of a MoleculeExperiment object

To assess the disk and memory size of molecule data objects, we used the public CosMx data corresponding to the “Lung9_Rep1” sample, with 26 275 891 molecules detected over 900 features. We assessed on disk file sizes for the transcript csv file as made available from the NanoString website, a Gzip compressed csv.gz version of the file, as well as MoleculeExperiment objects exported to disk via readRDS, either including all additional columns or only keeping essential columns. To assess memory sizes, we compared the two MoleculeExperiment objects to a data.frame generated by reading the aforementioned csv file. We quantified file and object sizes using the file.size and object.size functions, respectively, and reported these in megabytes.

2.3 Virtual dissection of mouse brain region

To demonstrate the interoperability afforded in MoleculeExperiment, we took the tiny subset of the Xenium data and loaded the morphology image tiff into napari (https://napari.org/). Using the shapes tool, we hand-annotated two regions corresponding anatomically to the granule cell layer and molecular layers of the dentate gyrus (Oh et al. 2014), and exported this file. We then read this file into R and used the dataframetoMEList() function to assign a new boundaries slot to the MoleculeExperiment object. We then visualized the annotated region and summarized molecule counts in these regions by first adding over counts for cells with centroids within the annotated regions, and second by summarizing directly on molecules using the countMolecules() function. We directly compared the per-gene counts between these regions in scatterplots.

3 Results

Here, we introduce MoleculeExperiment, a core data infrastructure package in R/Bioconductor which enables consistent and reproducible analysis of molecule resolution ST data in the R coding environment (Fig. 1A). The MoleculeExperiment class is an S4 class with one required slot for storing information on molecules and is nested by assay, e.g. for different transcript decoding approaches (Gataric et al. 2021; Cisar et al. 2023), by sample for datasets with multiple samples and/or images, and by feature_id for different transcripts or molecules. The core information in feature_id is the x_location and y_location of each molecule, but other additional information can be stored here (e.g. molecule-level annotations). The MoleculeExperiment object can contain an additional slot for storing boundaries, which is used for storing various segmentations of the data and is also nested by assay, for different segmentations such as cell bodies, nuclei, or annotated virtual dissections, by sample, and by segment_id for each individual segment (typically a cell).

Figure 1.

Figure 1.

Open in a new tab

MoleculeExperiment aims to facilitate molecule-level and cell-level analysis of data across different vendors of imaging-based ST data. (A) A MoleculeExperiment object has a molecules slot and a boundaries slot, where data format and terminology are standardized (e.g. hierarchical nested list for storage, µm units for the coordinates, and specific column names). The MoleculeExperiment class enables a molecule-centric analysis via class-specific accessor functions. (B) Possible molecule-level downstream analyses include visualizations (e.g. digital in situs), operations (e.g. filtering and counting), and spatial statistics [e.g. Differential gene expression (DGE) by virtual dissection]. In addition, the MoleculeExperiment package facilitates transition to a cell-level analysis via the summarization of molecule-level data into a SpatialExperiment object. (C) Screenshot of Xenium mouse brain tiny subset with hand annotation of two regions, in napari. (D) Visualization of annotated regions from MoleculeExperiment. (E) Comparison of total molecules detected for each gene in Region 1 (left) and Region 2 (right) showing aggregate over cell counts (x-axes) opposed to countMolecules over virtual dissection region (y-axes). Inset scatterplots are shown over entire dynamic range for all genes. Red dashed lines indicate line of best fit, and gray dotted lines indicate y = x line.

Due to the large variation in data bundles produced by various vendors of imaging-based ST technologies (Supplementary Figure 1), we have implemented specific functions for reading and standardizing this data into a MoleculeExperiment object. Currently implemented are readXenium(), readCosmx(), and readMerscope(), alongside a technology-agnostic dataFrameToMEList() function. The package provides setter and getter functions, e.g. molecules() and boundaries(), needed to manipulate the object in R. For segmentation information, we enable the reading of both boundaries via dataframeToMEList() as well as segmentation masks via readSegMask() functions.

The hierarchical nested structure of the MoleculeExperiment class avoids redundant storing of information, as opposed to traditional rectangular data storage formats (e.g. csv files). For example, the sample IDs and feature names are not repeated for the millions of molecules corresponding to that sample and feature. As such, MoleculeExperiment creates objects that consume less memory than rectangular objects (Supplementary Figure 2). Further, we provide the countMolecules() function that uses parallel computation to summarize the molecule-level data to a typical cell-by-gene representation in a SpatialExperiment object (Righelli et al. 2022). Thus, the MoleculeExperiment object not only takes advantage of the molecule resolution of imaging-based ST technologies, but also facilitates the transition from a molecule-level analysis to a cell-level analysis, thereby leveraging the vast capacity of Bioconductor tools designed for single-cell and spatial genomics analysis.

The molecule-centric way in which the MoleculeExperiment object stores data can be used for molecule-level visualizations and statistical analyses (Figure 1B), enabling wrangling of a large proportion of molecule data not overlaid onto existing cell segmentations. One crucial aspect to analysis of molecule-resolved ST is the ability to perform in silico virtual dissection and further analysis. We demonstrate the ease of use by virtually dissecting two regions of interest in the napari software (https://napari.org/) from the 10× Genomics Xenium Mouse Brain Tiny Subset data (Figure 1C, Methods) and jointly visualizing in R using MoleculeExperiment (Figure 1D). Summarizing over molecules in these regions results in higher recovery of gene counts compared to adding per-cell gene counts, in both dense and sparse tissue regions (Figure 1E).

4 Discussion

Here, we have developed MoleculeExperiment, an S4 infrastructure in R/Bioconductor that enables the analysis of imaging-based ST data at the molecule-level, thereby making full use of the molecule resolution that these technologies can achieve. It imposes a standardization of the data such that the data structure, and associated terminology, are consistent across data from diverse vendors of imaging-based ST. This consistent data representation aims to provide a solid foundation for the development of tools for the analysis of molecule-level ST data. We think this is especially important in the current context, where new imaging-based ST technologies and associated analytical methods are constantly being developed (Williams et al. 2022; Wu et al. 2022). The MoleculeExperiment package provides convenience functions to read data from different vendors of imaging-based ST technologies. This aims to simplify the otherwise manual and time-consuming process of in-house data wrangling before data analyses. Finally, the package facilitates summarization into a SpatialExperiment object for downstream cell-level analyses. In this way, it is possible to use already existing Bioconductor packages that work with the SpatialExperiment classes, e.g., imcRtools (Windhager et al. 2021), spicyR (Canete et al. 2022), SPIAT (Yang et al. 2020), and scHOT (Ghazanfar et al. 2020). Moreover, this means that the molecule-resolved data can be transitioned to other related classes, like SingleCellExperiment (Amezquita et al. 2020) and its python-equivalent AnnData (Virshup et al. 2021). Taken together, by being an S4 Bioconductor class, MoleculeExperiment profits from interoperability of downstream software packages, like the SPE (Righelli et al. 2022) and SingleCellExperiment (Amezquita et al. 2020) classes do. Ultimately, the MoleculeExperiment package imposes a consistent structure and terminology for imaging-based ST data, with the goal of enabling reproducible downstream molecular- and cellular-level analysis for the user.

Owing to its general nature, MoleculeExperiment could be applicable beyond imaging-based ST, and could be relevant to other novel technologies, e.g. lineage barcoding (Frieda et al. 2017), and indeed in any situation where discrete molecules are detected in spatial coordinates.

A key advantage of the nested structure of MoleculeExperiment is that it stores less redundant information in comparison to rectangular structures (e.g. csv files). Moreover, this hierarchical nested format enables parallelization via BiocParallel (https://bioconductor.org/packages/release/bioc/html/BiocParallel.html) (e.g. within the countMolecules() function). While current molecule-level ST datasets have few replicates, as these technologies increase to larger cohort scales, the need for on-disk representation of non-rectangular data will increase. Tools such as Apache Arrow (https://arrow.apache.org/) or the hierarchical data structure HDF5 (https://bioconductor.org/packages/release/bioc/html/rhdf5.html) may enable further development of on-disk representation of complex non-rectangular data, beyond arrays as used in other Bioconductor packages (Eling et al. 2021), that can be incorporated as classes within the MoleculeExperiment slots.

In summary, the MoleculeExperiment R package standardizes imaging-based ST data at the molecule level across different vendors, and simplifies the steps needed to prepare raw imaging-based ST data, ready for downstream analyses at the cellular- and molecular-level. We hope MoleculeExperiment supports the recent and fast-growing spatial omics community.

Supplementary Material

btad550_Supplementary_Data
Click here for additional data file. (625.5KB, pdf)

Acknowledgements

The authors thank all their colleagues, particularly at The University of Sydney, Sydney Precision Data Science and Judith and David Coffey Life Lab in Charles Perkins Centre for their support and intellectual engagement. We especially thank Nils Eling and Ludwig Geistlinger, along with all members of the Spatial Imaging Special Interest Group, for their careful feedback and engagement.

Contributor Information

Bárbara Zita Peters Couto, School of Mathematics and Statistics, The University of Sydney, Camperdown, NSW 2006, Australia; Charles Perkins Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Precision Data Science Centre, The University of Sydney, Camperdown, NSW 2006, Australia.

Nicholas Robertson, School of Mathematics and Statistics, The University of Sydney, Camperdown, NSW 2006, Australia; Charles Perkins Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Precision Data Science Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Laboratory of Data Discovery for Health Limited (D24H), Science Park, Hong Kong SAR, China.

Ellis Patrick, School of Mathematics and Statistics, The University of Sydney, Camperdown, NSW 2006, Australia; Charles Perkins Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Precision Data Science Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Laboratory of Data Discovery for Health Limited (D24H), Science Park, Hong Kong SAR, China; Centre for Cancer Research, The Westmead Institute for Medical Research, The University of Sydney, Westmead, NSW 2145, Australia.

Shila Ghazanfar, School of Mathematics and Statistics, The University of Sydney, Camperdown, NSW 2006, Australia; Charles Perkins Centre, The University of Sydney, Camperdown, NSW 2006, Australia; Sydney Precision Data Science Centre, The University of Sydney, Camperdown, NSW 2006, Australia.

Author contributions

SG, EP, conceived, designed and funded the study. BZPC completed the analysis and design of software with feedback from SG and EP. BZPC and NR implemented and constructed the R package with feedback from SG and EP. BZPC and NR tested the R package. All authors wrote, reviewed and approved the manuscript.

Supplementary data

Supplementary data are available at Bioinformatics online.

Conflict of Interest

None declared.

Funding

This research was supported by the AIR@innoHK programme of the Innovation and Technology Commission of Hong Kong to E.P.; Australian Research Council Discovery Early Career Researcher Awards (DE220100964, DE200100944) funded by the Australian Government to S.G. and E.P.; Chan Zuckerberg Initiative Single Cell Biology Data Insights grant (2022-249319) to S.G. The funding source had no role in the study design; in the collection, analysis, and interpretation of data, in the writing of the manuscript, and in the decision to submit the manuscript for publication.

Data availability

All data used in this study are publicly available. The accession links are reported in the Github repository at https://github.com/SydneyBioX/MoleculeExperiment.

References

  1. Amezquita RA, Lun ATL, Becht E. et al. Orchestrating single-cell analysis with bioconductor. Nat Methods 2020;17:137–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Canete NP, Iyengar SS, Ormerod JT. et al. spicyR: spatial analysis of in situ cytometry data in R. Bioinformatics 2022;38:3099–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Chen JH, Nieman LT, Spurrell M. et al. Spatial analysis of human lung cancer reveals organized immune hubs enriched for stem-like CD8 T cells and associated with immunotherapy response. bioRxiv 2023.04.04.535379, 2023, preprint: not peer reviewed. [Google Scholar]
  4. Cisar C, Keener N, Ruffalo M. et al. A unified pipeline for FISH spatial transcriptomics. bioRxiv, 2023.02.17.529010, 2023, preprint: not peer reviewed. [DOI] [PMC free article] [PubMed]
  5. Eling N, Damond N, Hoch T. et al. Cytomapper: an R/bioconductor package for visualization of highly multiplexed imaging data. Bioinformatics 2021;36:5706–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Frieda KL, Linton JM, Hormoz S. et al. Synthetic recording and in situ readout of lineage information in single cells. Nature 2017;541:107–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Gataric M, Park JS, Li T. et al. PoSTcode: probabilistic image-based spatial transcriptomics decoder. bioRxiv, 2021.10.12.464086, 2021, preprint: not peer reviewed.
  8. Ghazanfar S, Lin Y, Su X. et al. Investigating higher-order interactions in single-cell data with schot. Nat Methods 2020;17:799–806. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Hao Y, Stuart T, Kowalski MH. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol 2023. 10.1038/s41587-023-01767-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Janesick A, Shelansky R, Gottscho AD. et al. High resolution mapping of the breast cancer tumor microenvironment using integrated single cell, spatial and in situ analysis of FFPE tissue. bioRxiv, 2022.10.06.510405, 2022, preprint: not peer reviewed. [DOI] [PMC free article] [PubMed]
  11. Marconato L, Palla G, Yamauchi KA. et al. SpatialData: an open and universal data framework for spatial omics. bioRxiv, 2023.05.05.539647, 2023, preprint: not peer reviewed. [DOI] [PMC free article] [PubMed]
  12. Moffitt JR, Lundberg E, Heyn H.. The emerging landscape of spatial profiling technologies. Nat Rev Genet 2022;23:741–59. [DOI] [PubMed] [Google Scholar]
  13. Moore J, Basurto-Lozada D, Besson S. et al. OME-Zarr: a cloud-optimized bioimaging file format with international community support. bioRxiv, 10.1101/2023.02.17.528834, 2023. preprint: not peer reviewed. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Moses L, Einarsson PH, Jackson KC. et al. Voyager: exploratory single-cell genomics data analysis with geospatial statistics. bioRxiv, 2023.07.20.549945, 2023, preprint: not peer reviewed. [Google Scholar]
  15. Oh SW, Harris JA, Ng L. et al. A mesoscale connectome of the mouse brain. Nature 2014;508:207–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Palla G, Spitzer H, Klein M. et al. Squidpy: a scalable framework for spatial omics analysis. Nat Methods 2022;19:171–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Prabhakaran S. Sparcle: assigning transcripts to cells in multiplexed images. Bioinform Adv 2022;2:vbac048. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Righelli D, Weber LM, Crowell HL. et al. SpatialExperiment: infrastructure for spatially-resolved transcriptomics data in R using bioconductor. Bioinformatics 2022;38:3128–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Virshup I, Rybakov S, Theis FJ. et al. anndata: Annotated data. bioRxiv, 2021.12.16.473007, 2021, preprint: not peer reviewed.
  20. Williams CG, Lee HJ, Asatsuma T. et al. An introduction to spatial transcriptomics for biomedical research. Genome Med 2022;14:68. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Windhager J, Bodenmiller B, Eling N. An end-to-end workflow for multiplexed image processing and analysis. bioRxiv, 2021.11.12.468357, 2021, preprint: not peer reviewed. [DOI] [PubMed]
  22. Wu Y, Cheng Y, Wang X. et al. Spatial omics: navigating to the golden era of cancer research. Clin Transl Med 2022;12:e696. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Yang T, Ozcoban V, Pasam A. et al. SPIAT: an R package for the spatial image analysis of cells in tissues. bioRxiv, 2020.05.28.122614, 2020, preprint: not peer reviewed.
  24. Yao Z, van Velthoven CTJ, Kunst M. et al. A high-resolution transcriptomic and spatial atlas of cell types in the whole mouse brain. bioRxiv, 2023.03.06.531121, 2023, preprint: not peer reviewed. [DOI] [PMC free article] [PubMed]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

btad550_Supplementary_Data
Click here for additional data file. (625.5KB, pdf)

Data Availability Statement

All data used in this study are publicly available. The accession links are reported in the Github repository at https://github.com/SydneyBioX/MoleculeExperiment.

Articles from Bioinformatics are provided here courtesy of Oxford University Press

RESOURCES