Paste this into any page to embed the interactive figure. Optional URL parameters: color=class|subclass|ctxlen, ctx=500…8000, autoplay=1, panel=1 (legend open), caption=0 (omit the description, e.g. to use your own caption).
Bloom learns consistent representations across context sizes. A joint UMAP is fit on cell representations obtained with different input contexts. The input context for each cell consists of its n most expressed genes, with n ∈ {500, 1,000, 2,000, 4,000, 8,000}. The representation is remarkably robust over this 16× range of context sizes. Axes are the two UMAP dimensions. Each cell is matched across contexts by its ID; positions between measured contexts are interpolated for display.