Hejin0701
commented
1 year ago Hi Joy,
Yes. You should always train the model for new datasets. For your analysis, when you get a new single cell data and Visium data, you should first identify the marker genes. Then just follow the same steps in our tutorial but replace the single cell and ST data with your data (arranged in Anndata format). After that, use the map_cell_to_space function to train the model to get the mapping result.
Let me know whether that helps.
Best, Hejin
Hello I'm Joy,a graduate student, trying to use tangram. First, thanks for such a wonderful tool.
I was wondering that basic Tangram model already trained with mouse brain tissue only. I'm trying to segment cells(from other tissue which is not brain) with the cell segmentation function and proceed with deconvolution on visium data, because if the model is trained only with brain tissue, there is a possibility of analyzing it incorrectly. If so should i train the model? i thought that i don't need to train the model when proceed the deconvolution cuz the model already trained some data basically.
if i should train the model to handle other tissue, let me know how to train the model with a few tissue data and scRNA data.
Thank you. Joy