Reduced number of cell types in factorisation output

[georginaalbadri]

Hello,

I'm running tensor-cell2cell with liana and have done successfully for one of my datasets. For a second dataset, I have 7 cell types but only the same 3 are appearing in the sender cells/receiver cells in the factorisation results. What are the possible reasons for this? There are significant interactions (pval < 0.05) involving the other cell types in the liana output.

I just want to check if this is a possible output due to the nature of the factorisation, or if I might have to change one of the default inputs to make sure all cells types are included in the factors. I used elbow analysis so the number of factors should be suitable.

Thanks!

[dbdimitrov]

Hi @georginaalbadri,

It's likely that during the conversion of the liana results into a tensor, you have some cell types that are not shared across samples.

I suggest checking that when running liana you ensure you pass this parameter return_all_lrs=True. While when you convert the liana output into a tensor (via li.multi.to_tensor_c2c) you pass this one how='outer_cells'.

PS. I'm mostly guessing here, if you provide us with your code, I could give you better directions.