search

aertslab / AUCell

AUCell: score single cells with gene regulatory networks
122 stars 26 forks source link

determining thresholds #8

Closed ceesu closed 3 years ago

ceesu commented 4 years ago

Hello, thank you very much for your development of this package. I noticed that the signature thresholds that are automatically found for my single cell data are not always a great fit in the distributions of frequency vs. AUC. Do you think that it may be help for the input expression matrix to be normalized? If so how might this affect the auc scores and the final cell assignments?

s-aibar commented 3 years ago

The normalization of the input matrices should not affect much. The AUC is calculated for each cell individually. We have not found a better way to automate the thresholds, though. So they still need some user supervision...