Multiplexed RNA sequencing in individual cells is transforming basic and clinical life sciences1–4. Often, however, tissues must first be dissociated, and crucial information about spatial relationships and communication between cells is thus lost. Existing approaches to reconstruct tissues assign spatial positions to each cell, independently of other cells, by using spatial patterns of expression of marker genes5,6—which often do not exist. Here we reconstruct spatial positions with little or no prior knowledge, by searching for spatial arrangements of sequenced cells in which nearby cells have transcriptional profiles that are often (but not always) more similar than cells that are farther apart. We formulate this task as a generalized optimal-transport problem for probabilistic embedding and derive an efficient iterative algorithm to solve it. We reconstruct the spatial expression of genes in mammalian liver and intestinal epithelium, fly and zebrafish embryos, sections from the mammalian cerebellum and whole kidney, and use the reconstructed tissues to identify genes that are spatially informative. Thus, we identify an organization principle for the spatial expression of genes in animal tissues, which can be exploited to infer meaningful probabilities of spatial position for individual cells. Our framework (‘novoSpaRc’) can incorporate prior spatial information and is compatible with any single-cell technology. Additional principles that underlie the cartography of gene expression can be tested using our approach.
Bibliographical noteFunding Information:
Acknowledgements We thank A. Murray, A. Regev, T. Gregor, P. Rigollet, all members of our labs and many colleagues in the field for valuable comments and discussions. We thank L. Friedman for help with graphic design and illustration. This work was supported by the Israeli Science Foundation, through the I-CORE program (N.F.) and an Alexander von Humboldt Foundation Research Award (N.F.). N.K. was supported by grants DFG/GZ (Geschäftszeichen): RA 838/8-2 and DFG/GZ: KA 5006/1-1; and HGF Neurocure/GZ 0036-Phase 2-3. M.N. was supported by the James S. McDonnell Foundation, Schmidt Futures, the Israel Council for Higher Education and the John Harvard Distinguished Science Fellows Program within the FAS Division of Science of Harvard University. N.R. thanks Anna-Carina for useful discussions.
© 2019, The Author(s), under exclusive licence to Springer Nature Limited.