vitkl
commented
1 month ago In general, we recommend using as many samples as fit into the GPU memory because it helps cell2location distinguish cell abundance from technical sensitivity effects. So concatenate anndata, then apply cell2location.
You can read our paper methods sections for a detailed explanation of the multi-sample model and the benefits it provides.
In some cases, such as when samples have very different sources (eg Visium fresh frozen vs Visium FFPE), samples of different source would have to be analysed separately.
Zhongzheng99
hello, @vitkl . I have slides from three groups, and following previous tutorials (https://cell2location.readthedocs.io/en/latest/notebooks/cell2location_short_demo.html#1.-Loading-Visium-data), I've combined all the Visium data into one anndata object using concat, then estimated the posterior distribution of cell abundance. However, in the Scanpy official tutorial, they recommend analyzing the Visium data of each slide individually (https://scanpy.readthedocs.io/en/latest/tutorials/spatial/integration-scanorama.html#data-integration-and-label-transfer-from-scrna-seq-dataset). I'm curious about the best practice: should I integrate first and then perform spatial mapping?