This repository is an exploratory resource to accelerate opensource analysis of CosMx™ Spatial Molecular Imager (SMI) data. Contained here are and writeups and vignettes addressing a variety of topics discussed when analyzing single-cell spatial data.
different analyses have different solutions:
-- umap: use harmony or something like it.
-- DE: include batch as a variable
-- cell typing: leiden should consider harmony. Insitutype should be fine, but can also run separately on each batch (for supervised), allowing the platform calibration via anchors to work separately on each batch. Or else cell type one batch first, then use it as a reference for future batches.
-- QC: nothing needs to be done
-- normalization: you could apply scaling factors to individual genes, preserving a sparse matrix but moving from integers to doubles.
need to discuss what shape batch effects take - most countermeasures assume they're linear on gene detection efficiency
for advanced users: describe the insitutype platform scaling method, link to the relevant function
Notes to include: