search

hemberg-lab / scmap

A tool for unsupervised projection of single cell RNA-seq data
http://bioconductor.org/packages/scmap
GNU General Public License v3.0
90 stars 11 forks source link

Why are the clusters of the reference dataset changed? #15

Closed xbjie123 closed 5 years ago

xbjie123 commented 5 years ago

Hi,

I'd like this package very well. But here I have a little problem about the scmap-cluster.

When I did all the process for projection, the results were a little confused. The clusters are changed and the ratio of different clusters are quite dfifferent. Also, there are some unassigned cells. But as I know, the clusters would not change after the process, so I don't know what's the problem. default

Thanks very much and hope to get the reply as soon as possible.

BJ

wikiselev commented 5 years ago

Hello, from your screenshot it looks like your clusters may be quite weak and not very well separated. That's the reason when scmap assigns all to them to State I.

xbjie123 commented 5 years ago

Thank you very much for the quick reply. That's true. The clusters are not very well separated. So it means this method is not proper for these clusters? In fact, I want to observe the similarities of clusters from two different datasets. So I set one dataset as the reference (right one), the other one is the projection (left one). What I learned about this method is it can keep the identities of cells. Is it right? So I don't know why there are some unassigned cells and why the cell numbers in some clusters are decreased.

Thanks a lot. BJ

wikiselev commented 5 years ago

Your result suggests that the clusters are quite weak, it does not mean that scmap is not correct, it's just telling you what happens. Unassigned cells mean that their similarity to any of the clusters is lower than a threshold (0.7 by default, can be controlled manually).

xbjie123 commented 5 years ago

Got it. I also did the self-projection, however, there are still some unassigned cells. So does it mean that I could get the proper threshold when the self-projection results without unassigned cells?

Best BJ

wikiselev commented 5 years ago

No, the assignment result will be the same. With the lower threshold only the unassigned cells will be assigned to one of the clusters.

xbjie123 commented 5 years ago

OK. Thank you very much for everything.