reproduce deconvolution spatial trascriptomics

[bio-la]

Hi @roramirezf ! thanks for providing the reproducibility code for your paper!
I would like to use the cell2location deconvoluted cell types expression from the spatial data.

(

visium_heart/st_snRNAseq/04_deconvolution/.ipynb_checkpoints/run_c2l-checkpoint.py

Line 125 in 5b30c7e

inf_aver = adata_snrna_raw.raw.var.copy()

)

Could you share the trained cell2location models or alternatively the estimated celltype signatures?
thank you!

[roramirezf]

Hi thanks for approaching, the results of the deconvolution with cell2location can be obtained from the Seurat objects available here: https://zenodo.org/records/6580069#.YvET1-xBx-U or in the HCA portal https://explore.data.humancellatlas.org/projects/e9f36305-d857-44a3-93f0-df4e6007dc97

Hope this is helpful

[wxicu]

Hello, and thank you for sharing this information! Just to confirm, in the Seurat objects, c2l indeed stores the cell abundance, while c2l_prop saves the cell proportions?
If it’s not too much trouble, would you be able to share the c2l models as well? We’re hoping to obtain cell-type-specific gene expression, following the guidance here: cell2location documentation. Access to the model directly would be immensely helpful, if possible.

Thank you very much for considering this request!
Xichen

[wxicu]

Hi,
I am trying to rerun NB regression with cell2location. Could you please provide the batch information used in the script? I noticed a related discussion in issue #18, but I couldn't find any sample prefix in the h5ad file from the snRNA dataset downloaded from Zenodo (https://zenodo.org/records/6578047, snRNA-seq-submission.h5ad). Thank you!

https://github.com/saezlab/visium_heart/blob/5b30c7e497e06688a8448afd8d069d2fa70ebcd2/st_snRNAseq/04_deconvolution/nb_estimates_states_singularity.py#L88C25-L88C50

[roramirezf]

Hi Xichen,
At the time I built 6 iterations of the nb model with a subsampling of the data due to computational issues, and the mean estimates were the ones I used for final deconvolution. I would recommend to rerun the nb model for you applications.

I apologize for the batch information, if you are using the cellxgene data with CK ids, this table should contain the batch that you need.

snrna_batch_ann.csv

[wxicu]

Hi,

Thank you very much for providing the batch information; it’s really helpful. Another question: In the metadata for the Visium samples, there are two different sample IDs for patient P6, specifically Visium_17_CK295 and Visium_2_CK280. According to the clinical data, these samples have different infarction locations. However, in the final processed data shared on Zenodo, there is only one file for P6. Does this mean that the two samples for this patient have been merged?

Thank you in advance for your clarification!

Best regards,
Xichen

[roramirezf]

Hi Xichen,
Sorry for the misunderstanding. P6 was considered to be only remote zone of the MI patient, regardless of the sampling location that is reported in the Sup Table 1 of the paper which reports two different tissue samples (Visium_17_CK295, Visium_2_CK280), for the same reason there's only 1 dataset from single-nuc (CK356).

Another note, the c2l models that I fitted were done using the complete atlas rather than a 1 to 1 mapping between single-cell and visium, if that also helps to clarify a bit.

Cheers,
Rico