Question: How should I annotate the topics from reference free spot deconvolution?
Answer: The Visium Spatial Gene Expression solution generates spatial, whole-transcriptome data, but each spot typically covers multiple cells. To analyze data at cell type resolution, emerging spot deconvolution methods use single cell reference data to break down Visium spots into their respective cell types (reviewed in this 10x Analysis Guide).
Starting in Space Ranger v2.1 and Loupe Browser v6.5 take this a step further and introduce combined support for the reference-free spot deconvolution feature, a convenient alternative for when a single cell reference is not available, or to visualize cell-type data directly without relying on third-party tools. The algorithm is built on latent Dirichlet allocation (LDA).
Space Ranger's spot decovolution predicts latent topics that can be thought of as inferred cell types or mixtures of cell types. Familiarity with sample biology is valuable during the annotation process.
A tutorial for Loupe Browser is available on the support site.
Additional guidance in the following analysis guides articles using community developed tools.