Trajectory Analysis scRNA-seq with one cell type and two developmental stages

Question

[npont]

Hi Biostars community,

I would like to perform a trajectory analysis for a given cell type. I have two datasets (coming fro spatial visium HD, but let's consider them as scRNA-seq data) that I subset to select only cells corresponding to my cell type of interest. The two datasets represent two distinct developmental stages.

I was thinking of doing: 1) subset each dataset separately for my cell type of interest 2) merge datasets (without integration) 3) log normalize 4) perform UMAP to see if batch effect correction is required 5) diffusion map (palantir? cellrank? monocle?) to find the pseudotime and then once the pseudotime is chosen as a DC: 6) plot the expression of transcription factors that I found to be markers of my cell type against the pseudotime

Has anyone already done a similar analysis? I would like to make sure that this method makes sense, and especially the way I deal with my cell type (selecting only those cells as a first step) and the fact that I only have two developmental stages

Thanks a lot!

[Bastien Hervé]

I am not sure if there is a "best" approach to do what you want.

Some comments :

let's consider them as scRNA-seq data

The visium HD goes even lower than single cell resolution (2umx2um resolution) and after aggregation to 8um you cannot exclude that you are maybe covering 2 or 3 cells and not a single cell.

That said, for the analysis part I would first merge the datasets, look for batch effects across multiple cell types would be easier than only on your cell type of interest. Then, subset your cell type of interest and do 4, 5, 6)

Here they do some kind of pseudotime on Visium HD data

[npont]

Hello, Thank you for your reply

Indeed visium HD goes down to 2um resolution but I applied a nuclei segmentation which worked pretty nicely!

So do you think that the following workflow would make sense:

• merge both datasets (using concatenate function of anndata, with outer join)
• normalization
• umap to visualize the need for batch correction
• batch correction
• subset for my cell type of interest
• diffusion map & color cells by developmental stage to visualize which DC axis could be my pseudotime

Thanks for any comments that might help

[Bastien Hervé]

It makes sense to me to give it a try like this, however along the way depending on your batch effect correction or your pseudotime sanity check you would need some extra steps like reclustering your cell type.

PS : I see you're from France -- just out of curiosity, which platform in France has a VisiumHD ?

[npont]

Yes I would agree for clustering, but the way I define my cell type is not based on clustering: I define cells as being of my type of interest if they meet two conditions -> being located in a manually defined region corresponding to my muscle of interest and expressing two known markers specific to this muscle. Let's see how it goes!

Concerning the visiumHD platform, it was done at the institute where I work which is a public research institute whose name I can tell you in DM if you want