Inferring spatial and signaling relationships between cells from single cell transcriptomic data.
Cang Z, Nie Q. · 2020
Read Inferring spatial and signaling relationships between cells from single cell transcriptomic data. by Cang Z, Nie Q. (2020) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc7190659.txt.
Introduction
Cang and Nie present a method that restores spatial information to single-cell gene data. Their study appeared in Nature Communications. Single-cell sequencing reveals each cell's activity in fine detail. But the process destroys the cell's original location in the tissue. The method, called SpaOTsc, rebuilds that lost geography.
It uses spatial measurements of a small set of genes as an anchor. A mathematical transport approach then places each cell on the map. From those positions, the method infers which cells signal to each other. It also estimates how genes in different cells influence one another. Tests on four datasets checked the predictions. Work on fly embryos recovered known signaling patterns. Biologists gain a way to study tissue conversations in place.
Catalog record
| CREATOR | Cang Z, Nie Q. |
|---|---|
| YEAR | 2020 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Inferring spatial and signaling relationships between cells from single cell transcriptomic data By Zixuan Cang, Qing Nie Published in: Nature Communications DOI: 10.1038/s41467-020-15968-5 ABSTRACT Single-cell RNA sequencing (scRNA-seq) provides details for individual cells; however, crucial spatial information is often lost. We present SpaOTsc, a method relying on structured optimal transport to recover spatial properties of scRNA-seq data by utilizing spatial measurements of a relatively small number of genes. A spatial metric for individual cells in scRNA-seq data is first established based on a map connecting it with the spatial measurements. The cell–cell communications are then obtained by “optimally transporting” signal senders to target signal receivers in space. Using partial information decomposition, we next compute the intercellular gene–gene information flow to estimate the spatial regulations between genes across cells. Four datasets are employed for cross-validation of spatial gene expression prediction and comparison to known cell–cell communications. SpaOTsc has broader applications, both in integrating non-spatial single-cell measurements with spatial data, and directly in spatial single-cell transcriptomics data to reconstruct spatial cellular dynamics in tissues. INTRODUCTION Single-cell transcriptomics methods enable analyses of gene expression heterogeneities in individual cells to study cell fate decisions1. Dissociation of tissues into single cells allows high-throughput genomics measurements, but spatial information of cells is often lost. While single-cell transcriptomics has mainly been used to delineate cell subpopulations and their lineage relationships, recently computational tools have also been developed to infer cell–cell communications from scRNA-seq data2,3. For example, by comparing the average enrichment of genes involved in different cell subpopulations, one might describe the signaling activities of each subpopulation4. A probability model that correlates ligand and receptor (and the downstream genes) expression levels in different cells allows the inference of communications between individual cells5. At the level of cell populations, a similar approach based on known ligand–receptor pairs was used to derive communications between cell types6, which can be further refined using prior knowledge of cell types7. Using cell type-specific enrichment of genes in known gene regulatory networks, one can also infer communication among cell clusters8. Despite rich details on genes contributing to cell–cell communication in scRNA-seq data, the lack of spatial information in such datasets restricts its usefulness for studying cell–cell communication in tissues with spatial structure. On the other hand, measuring gene expression in intact tissues provides spatial resolutions but the genes examined need to be selected in advance. Is it possible to better infer communications between cells located in different positions in the intact tissues using single-cell transcriptomics data with the aid of additional spatial measurements? Several methods have been developed to pair scRNA-seq data with spatial information using spatial imaging data (e.g., in situ hybridization). For example, spatial information was obtained at cell cluster levels by identifying spatial domains with coherent gene expressions in spatial imaging data combined with scRNA-seq data9. At an individual cell level, similarity measurements based on correlation coefficients10,11 or correspondence scores12 between commonly examined genes in both spatial imaging data and scRNA-seq data were used to reconstruct spatial gene expression or map cells in scRNA-seq data to their potential spatial origins. Posterior probability estimates were carried out on spatial data described by a mixture model13 or simplified to one-dimensional bins14 to assign spatial origins to individual cells. Other general methods designed for the integration of multi-omics data can also be applied to integrate these two data types. Canonical correlation analysis was used to connect cells in scRNA-seq data to locations in spatial data15, facilitating the subsequent identification of anchors across the datasets for integration16. Non-negative matrix factorization can also be used to construct common low-dimensional spaces of multiple datasets17. These methods connecting scRNA-seq data and spatial data are especially valuable for analyzing spatial patterns of different genes or cell types in embryos10,13 and organisms with robust patterns9,12,14. These existing approaches focus on reconstructing spatial gene expression or estimating the spatial origins of cells in scRNA-seq data. The heterogeneity in single-cell measurements might be averaged out when determining the spatial gene expression patterns since multiple cells can be mapped to a single position. There also might be multiple highly possible spatial origins for an individual cell in scRNA-seq data. This makes existing approaches difficult to incorporate with cell–cell communications. Here we utilize the optimal transport method18 to equip cells in scRNA-seq data with a spatial distance with single-cell resolutions by connecting with another dataset on spatial measurements of a small number of genes. Optimal transport allows natural coupling of distributions (pairing of datasets) and characterization of distances between multiple distributions (the difference between datasets or data samples represented as distributions)18. Recent advancements in optimal transport method development, including efficient algorithms19, accessible library20, and flexible formulations21,22, enable its broader application23–25, such as inference of developmental trajectories26 and handling batch effects27 in scRNA-seq data. To connect individual cells and the spatial positions in two different measurements, we first develop a map between the two datasets by spatially optimal transporting the single cells (SpaOTsc) to spatial imaging datasets. The cell–cell distance calculated from the SpaOTsc map yields a spatial metric for scRNA-seq data. We then use this metric to establish an optimal transport plan from a probability distribution of “sender cells” to “receiver cells.” Such sender cell distributions can be characterized by the expression levels of communication genes (e.g., ligands) while the receiver cells can be distinguished by the paired genes (e.g., receptors and ligand–receptor downstream genes). Our approach mimics the corresponding physical processes of ligand release by signal senders and consumption by potential receivers. After obtaining an initial cell–cell communication network, we then use a machine learning model based on ensemble of trees to estimate the spatial range of signaling for spatial cell–cell communications. While cells may communicate with each other directly through ligand–receptor interactions, a gene in one cell may affect another gene in other cells indirectly. To explore influences among genes across different cells, which may not be directly interacting in cell–cell communications, we then use partial information decomposition28,29 to quantify the unique information provided by one gene to another gene across different cells. For a given pair of genes and a prescribed spatial distance, we quantify how likely they are interacting across different cells through space, referred to as intercellular gene–gene regulatory information flows. We first test SpaOTsc through cross-validation of spatial gene expression predictions as well as spatial mapping of single cells with known origins in four pairs of single-cell RNA-seq and spatial gene expression measurements. We then infer cell–cell communications and intercellular gene–gene regulatory information flows for three systems. RESULTS Overview of SpaOTsc method SpaOTsc method consists of two major components: (a) constructing a spatial metric for cells in scRNA-seq data and (b) reconstructing cell–cell communication networks and identifying intercellular regulatory relationships between genes (Fig. 1). A spatial metric for the cells in scRNA-seq data is first constructed using a mapping to spatial data. Using this spatial metric, we generate spatial visualization and clustering of cells and genes for scRNA-seq data. Cell–cell communication networks are then reconstructed for particular signalings. Finally, by feeding the spatial metric and scRNA-seq data to machine learning models and partial information decomposition, we infer the spatial ranges of particular signalings and quantify intercellular gene–gene regulatory information flows between genes. See more details in Methods