Add function to transfer cell type labels in MappingProblem and memory allocation issues

[valelorenzi9]

Hi,
Thank you very much for developing this framework! I am using the MappingProblem to transfer gene expression information from scRNA-seq data to In Situ Sequencing data and I have two questions:

1. What are the memory requirement for the mp.impute() and mp.correlate() functions? For mp.impute(), if I try to impute all the HVGs in my scRNA-seq dataset (~3k genes) I get XlaRuntimeError: RESOURCE_EXHAUSTED: Out of memory allocating XXXX bytes. If I specify a subset of ~30 genes to impute then it works perfectly fine, but I was wondering what is the upper limit of genes you can impute? The same memory error is also thrown when running mp.correlate(). I am using a VM with 300GB of RAM.

2. Are you planning to implement a function to also transfer the cell type labels from the dissociated data to spatial data or if not do you recommend a specific way of approaching the problem using the transferred gene expression information?

Thank you very much in advance!
Valentina

[giovp]

hi @valelorenzi9 , thank you for your interest in moscot.

What are the memory requirement for the mp.impute() and mp.correlate() functions? For mp.impute(), if I try to impute all the HVGs in my scRNA-seq dataset (~3k genes) I get XlaRuntimeError: RESOURCE_EXHAUSTED: Out of memory allocating XXXX bytes. If I specify a subset of ~30 genes to impute then it works perfectly fine, but I was wondering what is the upper limit of genes you can impute? The same memory error is also thrown when running mp.correlate(). I am using a VM with 300GB of RAM.

it should work for more than 30 genes, although to be fair we don't have the batch_wise implementation as it is done in cell_transition. Couple of questions on this:

• are you running it on GPU? if yes, try to pass device="cpu" to the function call
• how many cells are in the source and target distribution?
• are there batches in the spatial data? meaning, do you explicitly pass batch_key in prepare?

an easy solution to this would be to do a for loop and concatenate resulting the anndatas such as:

adatas_l = []
for genes in np.array_split(adata_sc.var_names, 100): # split all genes in 100 lists of ~30 genes
    adatas_l.append(mp.impute(var_names=genes,))
adata_imputed = ad.concat(adatas_l, axis=1)

Are you planning to implement a function to also transfer the cell type labels from the dissociated data to spatial data or if not do you recommend a specific way of approaching the problem using the transferred gene expression information?

you could use the cell_transition method or alternatively this:

dummy = pd.get_dummies(adata_sc.obs["annotation"])
out= mp[("src", "tgt")].pull(dummy, scale_by_marginals=True)
clusters = pd.Categorical([dummy.columns[i] for i in np.array(out.argmax(1))])
adata_spatial.obs["annotation_mapped"] = clusters

the difference between the two methods is the way the cluster assignment for a spatial cell is selected, the first it selects base on sum of the transportation cost, the second based on the argmax. The former is more conservative and might return fewer clusters than the ones in the source. That might be a sensible thing (especially in the non-low rank case where you might have explicitly set some tau for the unbalance case) but might also not. Interpretation would be required.

[valelorenzi9]

Thank you very much for the super helpful answers, @giovp!
To your questions:

• As this is a small spatial sample, I am running everything on CPU
• I have 34424 cells in the source (single cell) and 8391 cells in the target (spatial) distribution
• There are no batches in the spatial data, it is just one sample/image, so I am not passing the batch_key in prepare

For now your for loop solution worked perfectly fine and in very reasonable time, thank you for that!

Would you be able to provide an example of how to transfer cell type annotations with the cell_transition method? I don't fully understand how to make it work in the case where you don't have any annotations in the spatial data.

Thanks again!!

[giovp]

hi @valelorenzi9 , apologies for late reply, you can try something like this

df = problem.cell_transition(
    source="src",
    target="tgt",
    target_groups="annotation",
    forward=True,
    aggregation_mode="cell",
    batch_size=1024,
)

if it asks for source_groups then pass any adata.obs column, effectively it's not used but we should probably change that behaviour.

I have 34424 cells in the source (single cell) and 8391 cells in the target (spatial) distribution

understood. This means that you can play around with tau_a and tau_b to make the problem unbalanced and hence potentially improve results by "not transporting" outliers. let me know if that helps

[giovp]

implementation idea for this is the following @ArinaDanilina

one method called celltype_mapping of SpatialAnalysisMixin that accepts two modes: sum is the cell_transition, max is the argmax of the pullback distribution (see comment above). Also all the arguments of cell_transition or pullback needs to be exposed, as well as the potential target name (e.g. tgt by default, but if user has batches, then they should also be passed).

[ArinaDanilina]

hi @valelorenzi9 , we now have an example showcasing the function for annotation label transfer
https://moscot.readthedocs.io/en/latest/notebooks/examples/problems/900_annotation_mapping.html