Difference between is_doublet and preds

[ymahmoud]

Hi!
I'm trying your tool to identify doublets in my scRNASeq data but I'm not sure why I get slightly different results in is_doublet.csv and preds.csv files. I'm not using the -e parameter (expected_number_of_doublets), so shouldn't they be the same? What's the difference between these files?

Thanks!!
Yamil

[njbernstein]

Great question and something I need to clean up in the readme. They are supposed to be different now because we recalibrate the predictions the come directly out of solo to account for the fact that during training the model sees twice as many doublets as singlets. Because this doesn't reflect what a real scRNA-seq experiment will look like the scores get recalibrated and then cells with a softmax greater than .5 get called a doublet. This is what is written to is_doublet.csv and is the file you want. preds.csv are the raw predictions out of solo

Long story short is_doublet.csv are the doublet calls you want.

[ymahmoud]

Great question and something I need to clean up in the readme. They are supposed to be different now because we recalibrate the predictions the come directly out of solo to account for the fact that during training the model sees twice as many doublets as singlets. Because this doesn't reflect what a real scRNA-seq experiment will look like the scores get recalibrated and then cells with a softmax greater than .5 get called a doublet. This is what is written to is_doublet.csv and is the file you want. preds.csv are the raw predictions out of solo

Long story short is_doublet.csv are the doublet calls you want.

Great, I got it. Thank you for your quick response!