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

Quite low accuracy using index cluster #38

Open Kevis9 opened 2 years ago

Kevis9 commented 2 years ago

Hi, I am trying to use scmap to predict unlabeled datasets (let's call query) based on reference dataset. I use scmap index cluster followed by this vignette. The prediction accuracy is quite low, about only 0.007. However accuracy produced by scmapCellindex is quite high, about 0.91. I checked other papers which also used scmap to predict labels, it seems like indexCluster function work well at their experiment. I don't know whether there is any mistake in my code.

I checked prediction results, there lots of "unassigned" label in it. Can you give an example of how to use scmap index cluster to do prediction work based on reference datasets projecting to query dataset ?

mhemberg commented 2 years ago

Not sure if you have already tried this, but the best way of getting rid of "unassigned" labels is to lower the "threshold" parameter (which exists for both versions of scmap). Not really sure why one method does not run into this problem, but I would suggest gradually lowering it from the default of 0.7.

Hope this helps.

Kevis9 commented 2 years ago

Thank you. That's very helpful.