search

mahmoodlab / HEST

HEST: Bringing Spatial Transcriptomics and Histopathology together - NeurIPS 2024
Other
164 stars 12 forks source link

Image registration in Xenium dataset #45

Closed xiachenrui closed 2 months ago

xiachenrui commented 2 months ago

Hi all,

Thanks for your great works!

I want to know if H&E images in Xenium has been registered with cells coordinates. Because in the raw experiment they are not registered. 10x Xenium analysis pipeline provides a file (such as Supplemental: H&E Image Alignment File (CSV) ) for registration. Do you have applied this file to image ?

Any clarification will be helpful !

Best!

xiachenrui commented 2 months ago

I have this question because I found that the cell coordinates kept in adata.obsm['spatial'] can not matched well with cell location segmented via CellViT (kept in {name}_cellvit_seg.geojson file)

pauldoucet commented 2 months ago

Hi @xiachenrui,

See Issue #38, we're currently working on providing .h5ad with transcripts pooled per cell.

adata.obsm['spatial'] in st/*.h5ad does not contain cell coordinates. For now, adata.obsm['spatial'] in the provided .h5ad, only indicates the center of pseudo-visium spots for Xenium, which are pooling of transcripts per region of 100um x 100um.

To answer your question, no, we do not apply any transformation to the H\&E image directly as it's too costly. However the pseudo-spots themselves were indeed aligned using an affine alignment file. The CellViT segmentation is also already aligned with the H\&E by design.

Let me know if it's not clear

xiachenrui commented 2 months ago

Thanks for your quick reply ! My question is only about Xenium dataset, sorry for misleading.

For Xenium slices, I think adata.obsm['spatial'] is coordinates for cells defined by 10x xenium analysis pipeline. Does these coordinates already aligned with H&E (or CellViT results)?

Best!

pauldoucet commented 2 months ago

Sorry, can you clarify which adata you are talking about? Is is the one you obtained from our huggingface page i.e. st/....h5ad?

guillaumejaume commented 2 months ago

In Xenium, coords provided as part of HEST in adata.obsm['spatial'] are not cells, but patches, where transcripts have been pooled to emulate a Visium spot. We are in the process of releasing the H&E-aligned Xenium coords for each cell.

xiachenrui commented 2 months ago

In Xenium, coords provided as part of HEST in adata.obsm['spatial'] are not cells, but patches, where transcripts have been pooled to emulate a Visium spot. We are in the process of releasing the H&E-aligned Xenium coords for each cell.

Thanks for your clarification and efforts ! I see !

Best!