search

drieslab / Giotto

Spatial omics analysis toolbox
https://drieslab.github.io/Giotto_website/
Other
257 stars 98 forks source link

cellcell interaction question #684

Open chrkuo opened 1 year ago

chrkuo commented 1 year ago

HI Giotto team,

In the current protocols paper - the downstream giotto analysis comprised of cellcell interaction -- this is built upon HMRF inferred data from deconvoluted spatial samples correct? so is this at a resolution of near single cell? since the data itself is based on the scRNAseq data? or is that the wrong assumption?

RubD commented 1 year ago

Hi @chrkuo In the current protocols paper it is not based on the HMRF results. Hence the results can not be understood at the level of single cells.

chrkuo commented 1 year ago

@RubD

the cluster column stated here (image shown below from the current protocols paper) is based on the HMRF results (at least that's my understanding of it) and it seems like the cellcell interaction/ligand-receptor analysis can't proceed until HMRF analysis is done. It seems like the whole workflow is contingent on the protocol stated in step 6 which is the HMRF analysis. Can you please clarify?

If it is based on HMRF results - does this mean that it can then be understood at the level of single cells?

image

https://labs.icahn.mssm.edu/yuanlab/wp-content/uploads/sites/372/2022/07/Giotto_CP_part2.pdf

Appreciate your help.

chrkuo commented 1 year ago

Hi @RubD wanted to follow up on the response. Thank you so much for your assistance.

Additionally, are there more up to date vignettes or tutorials regarding LR and cell-cell interactions with visium spatial data

gcyuan commented 1 year ago

HI @chrkuo , sorry about the confusion. As @RubD mentioned, the ICG analysis is independent of HMRF and its calculation requires the data has single-cell data resolution. The mathematical detail is described in our original Giotto paper (PMID: 33685491). In the Current Protocol paper you mentioned, we primarily used a Visium dataset to illustrate Giotto, which does not have single-cell resolution. Therefore, strictly speaking we cannot apply the same procedure. As a work-around, we made a few approximating assumptions. The first is that each cluster can be approximately interpreted as a cell type, and the second is that the cellular environment can be approximated by HMRF output. Under these assumptions, essentially we are just looking for differentially expressed genes between spatial domains, which is not exactly the same as the ICG definition as for single-cell data analysis. Please also note that the screenshot you copied is not Fig. 25, which is actually a dot-plot and located above the text you cited. Hope this clarifies your confusion. We are still working on how to carry out ICG and LR analyses for multi-cell resolution datasets.