search

kstreet13 / slingshot

Functions for identifying and characterizing continuous developmental trajectories in single-cell data.
265 stars 43 forks source link

slingshot not robust on UMAP #143

Closed Tianran1 closed 3 years ago

Tianran1 commented 3 years ago

Hi,

I use slingshot on the UMAP plot. When I change the min.dist of UMAP, the output of slingshot changes as well. The trajectory paths are different. And when I change the number of cells in some clusters by downsampling, the output lineage path stays the same, but the pseudotime changes, showing a quite different line on the UMAP plot. Do you have any ideas about this?

Best, Tianran

kstreet13 commented 3 years ago

Hi Tianran,

I can't say anything specific about your results without seeing the plots, but in general, I don't think this is too surprising.

Changing the min.dist (or any other UMAP parameter) affects the eventual output of UMAP, which fundamentally changes the space in which Slingshot is working. So it is not surprising that this would also change the Slingshot results (as we talked about previously, you might be able to compare Slingshot results across different spaces, but it takes some effort).

The downsampling results, however, are a bit concerning. I would expect Slingshot to be fairly robust to this sort of change. What do you mean when you say that the "lineage path" stays the same, but the pseudotime and "lines" change? It is perhaps worth pointing out that the units of pseudotime are not meaningful, so something like Spearman correlation would be an appropriate way to compare multiple sets of pseudotime values over the same cells.

Best, Kelly