Multi-slice analysis of deconvolved Visium data without ground truth regions/celltypes

[mbaardwijk]

Hi!

I would like to use your methods to investigate spatial niches across multiple slices of Visium FFPE data on which I have applied deconvolution using Cell2Location to identify cell-type compositions.

I have followed your tutorial on the DLPFC dataset, but I struggle to apply the same steps to my own data. The main issue is in the sn.pp.process_multi_slices function, where a celltype_key is asked. For the DLPFC dataset, the domains (e.g., WM, layer_1, layer_2, etc.) are used, but in my dataset, I do not have this type of ground-truth region. What would you advise using instead?

Thanks in advance for your help!

[ProDong0512]

Hi!

I would like to use your methods to investigate spatial niches across multiple slices of Visium FFPE data on which I have applied deconvolution using Cell2Location to identify cell-type compositions.

I have followed your tutorial on the DLPFC dataset, but I struggle to apply the same steps to my own data. The main issue is in the sn.pp.process_multi_slices function, where a celltype_key is asked. For the DLPFC dataset, the domains (e.g., WM, layer_1, layer_2, etc.) are used, but in my dataset, I do not have this type of ground-truth region. What would you advise using instead?

Thanks in advance for your help!

Sorry for the late reply. The param celltype_key is only used in the preprocess, but not used in the model training. For a simple running, you can simply define celltype_key the same as the param sample_key, which leads to a identical input and output.

For example, if your sample_key = 'slice', you may simply define celltype_key = 'slice'.

I hope this would help! And if there is any problem, please feel free to add!