search

kstreet13 / slingshot

Functions for identifying and characterizing continuous developmental trajectories in single-cell data.
258 stars 42 forks source link

slingshot analysis on PCA but visualization on UMAP #234

Closed TdzBAS closed 6 months ago

TdzBAS commented 8 months ago

Hi,

I am using slingshot for TI. I am wondering If I can use PCA as reducedDim argument in the slingshot function and for visualization the UMAP in embedCurves? Since I am getting biologically more reasonable results, if working in the PC space, because UMAP can distort data quitw a lot.. Is this a reasonable approach?

Thanks!

Cheers, Tolga

kstreet13 commented 8 months ago

Hi @TdzBAS,

Yes, I think this is a great approach! Running slingshot on the PCs and then embedding it in the UMAP space via embedCurves for visualization is generally my recommended approach.

Sometimes the UMAP distorts things too much and the results look a little odd. In that case, I would just recommend creating multiple small plots (one for each lineage) and coloring the cells by pseudotime along each lineage.

Best, Kelly

TdzBAS commented 8 months ago

Hi @kstreet13,

thanks four your quick and valuable answer!

So two more question, where you maybe can help me out:

  1. in a second dataset the Lineages and Pseudotimes look biologically more meaninful using UMAP for the analysis instead of PCA. Can I use the UMAP here or is this cherrypicking?

  2. I integrated a dataset using harmony. Should I do the analysis then on the harmony principal components?

Best, Tolga

kstreet13 commented 8 months ago

Sure thing! For your questions:

  1. That does sound a bit like cherrypicking. If I was a reviewer, I'd be a little suspicious of that.
  2. I think that would be the right choice, yes. I don't have much experience using Harmony, but I typically use the "corrected" PCs from fastMNN and I think that's basically the same idea.

Best, Kelly