- Big difference between "expression data for cells from which you know the location vs having coordinates of gene expression as derived by e.g. fluorescence techniques (e.g. seqFISH+, ) where you have subcellular resolution (and may or may not know the boundaries of cells in that coordinate space, here it might not make sense to have a genes x cells matrix)."
- "Standard representations of the processed data (spatially-annotated gene expression matrices) are the biggest need and the easiest place to find commonality across the many assays." (Justin Kiggins)
- "Subcellular features are important, at the very least for inspection and QC for the gene expression matrices. Our experience with starfish indicates that it should be possible to standardize the decoded spot calls across the image-based transcriptomics methods" (Justin Kiggins)
- "annotating x/y/z coordinates of a tissue sample onto a gene expression matrix is pretty straightforward. we also have a sense that there are other spatial features that biologists analyzing these data want, like summary descriptions of cell morphology (size, shape, etc), quality score, etc. Then, as you aggregate across multiple tissue samples, x/y/z becomes meaningless & you need secondary spatial labels, like region labels or common coordinate framework. the details here will depend on tissue and species norms" (Justin Kiggins)
- Andrew Jaffe: "first i dont think its a good idea to assume any of the spatial transcriptomics data is at single cell resolution, particularly not with the commercially-available visium only have 55um spot resolution. other approaches like slideseq and emerging approaches could also be subcellular resolution, like several micron which would also not result in anything at the single cell level. it would probably be easier to extend existing single cell objects to just have a spatial barcode as the sample instead of cell, and then try to have addition phenotype/sample information. second, you probably want some kind of container for the histology image itself, particularly for visium. then the sample/barcode information can have coordinates to overlay on the histology. its also possible to count how many cells are in each spot/at each barcode although its not super straightforward, which you could append to the sample data...this would likely be computed outside of R since R isnt great with image analysis, although maybe that could change over time. third, i also think it also might be hard for the microscopy-based approaches and sequencing-based approaches to have common data formats since the data are pretty different. the former are more likely to actually be single cell resolution since there are likely processing steps that result in a cell by gene matrix with spatial information that i assume would be done way before reading the data into some bioc class".