Integrate multiple VisiumHD datasets

[kebbekny]

Hi all and thank you for a great tool!

I am currently trying to merge and compare two datasets of VisiumHD (treatment vs control) and wonder what the appropriate way to do this is.

I have been following the Sketch-based assay but that is only for a single sample.

Q. Can I apply the sketch-based approach after merging the samples?

There is also this vignette which uses the Standard Definition Visium Workflow for merging and SCTranform normalization while the Sketch-based assay uses NormalizeData and FindVariableFeatures.
https://satijalab.org/seurat/articles/spatial_vignette.html

Confused as what is the appropriate way to proceed with confidence.

Appreciate any input.

[AHK42]

So far I have been using SCTransform and then integrating with harmony for my treatment vs control samples. Not sure if this the correct workflow, but my thought process is to cluster shared and unique cell types of both samples together, and then overlay those clusters on both images.

[kebbekny]

Dear @AHK42,

Thank you for your reply. This make sense. I am just afraid of over-correcting using integrations such as harmony as these methods are usually applied to account for batch-effect.

Therefore, I wonder if it is appropriate to proceed without integration and only apply normalization after merging samples that were run in the same batch.

Thank you for further input.

[AHK42]

From my understanding, Visium will show very strong batch effects from sample to sample, even in biological replicates. You can also try multiple integration methods and see what works better.

[kebbekny]

That makes a lot of sense. Indeed, I have noticed quite a lot of variability, even between samples within the same batch. Thanks for your comments.

[pikapika505]

I am going to integrate 4 HD Visium samples (2ctrl/2treatment). My plan is to merge all suerat objects with merge() and then do SCTransform and integrate either with CCA or harmony. In my understanding, we can treat bins as cells in scRNA-seq analysis and use "spatial.008um" as "RNA" assay. I will have to correct for batch effects as there will be technical batch effects.

[kebbekny]

Hi @pikapika505,

Thanks for your reply. Are your batch-effects due to the fact that your samples where processed independently during the VisiumHD assay or are there other technical batch-effects that I am not accounting for if the samples were placed on the same slide and sequenced together?

Thanks again.

[kebbekny]

Furthermore, conducting the CCA integration on the full datasets takes an incredible amount of computing power and may not be a feasible approach. Any suggestions?

[pikapika505]

Well, on second thought, I guess in my case I don’t need to integrate the samples. My goal is to find differentially expressed genes between cancer cells in two conditions. Since I’m confident about my cancer cell labels in each sample, I think I can skip integration and just proceed with pseudobulk analysis.

That said, I did merge the samples and run SCTransform earlier without fully thinking through my analysis steps - it took quite a while, especially with 700,000 bins in total.