This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Cell-cell communication for Visium HD

Hello,

I am trying to find a good method for finding cell-cell communications in Visium HD data. My data has ~140 000 segments (cells), I did cell segmentation with cellpose and treat each polygon as a cell.

My goal is to find ligand-receptor interactions based on gene expression and distances between cells, preferably in cell types agnostic way.

So far, I tried these methods:

  1. CellNEST - I could not reproduce their tutorial with the test data. It found way too many components that it was impossible to interpret.

  2. COMMOT - kept crashing. In their github I saw that some people had the same problem when applying it on VisiumHD. No response, I don't think they have a capacity for Visium HD.

  3. SpatialCellChat - also keeps crashing, the tutorial is written for Visium, so I don't know if they have capacity for HD data.

There is squidpy method but I don't think the distance is not accounted there.

So, I would be happy to get some advice from the community? What have you used for Visium HD analysis? Or is there a way to apply aforementioned methods on Visium HD?

visium spatial visiumhd omics
• 718 views • link
updated by OmnibusX • OmnibusX • written by fifty_fifty • fifty_fifty

You may want to take a look at Liana if you have no already done so.

• 0 views • link
written by mbyvcm mbyvcm

2 answers

as mbyvcm mentioned, check out LIANA+. They compile previously existing CCC methods and databases into one, which is pretty neat. You can either run their consensus method which sort of combines results from all available methods in their ecosystem. There's this newer tool called LARIS (https://www.biorxiv.org/content/10.1101/2025.11.26.690796v1) which seems pretty interesting (however I've yet to test it out yet), as it can compute bivariate LR specificity as well as infer directionality with their bivariate LR expression metric. I would love to hear more from you as you take on this analysis, because I have been struggling quite a lot with LR inference with VisiumHD data. The counts are so sparse so most magnitude metrics are unflattering. Personally, I have been taking a spatial/niche selection + pseudobulking method but even magnitude metrics get killed by sparsity when conducting analysis in that manner as well.

Also one note about the crashing issues you are having with tools you've tried so far, that's unsurprising with VisiumHD since it is much higher res than Visium V1. A lot of methods were made with Visium V1 and scRNA seq in mind, so they often struggle to scale to HD res.

• 0 views • link
written by danielpintard • danielpintard

I tried CellPhoneDB on a 10x Visium HD demo dataset, and in my case the results looked reasonably interpretable. enter image description here

What worked better for me was using the 16 µm bin level rather than pushing to the finest possible resolution, because it helps cover neighboring cells and makes the ligand-receptor signal a bit more stable in this very sparse setting.

I also wrote a short explanation of the workflow and how to interpret the results here, in case it is helpful: https://blog.omnibusx.com/dive-with-visium-hd-uncover-spatial-cell-communication-at-single-cell-clarity/

• 0 views • link
written by OmnibusX • OmnibusX

Log in to answer this question.

More posts like this