Open seb951 opened 1 year ago
Here is a minimal example to illustrate my point. I mean results are pretty similar in this example, but not the same. Any reason why this is or which one to trust?
#setup
packageVersion(pkg = "Seurat")
#[1] ‘4.3.0’
packageVersion(pkg = "Azimuth")
#[1] ‘0.4.6’
SeuratData::InstallData('pbmc3k')
pbmc3k = SeuratData::LoadData('pbmc3k')
SeuratDisk::SaveH5Seurat(pbmc3k,'pbmc3k.h5Seurat')
#run Azimuth locally
pbmc3k = RunAzimuth(pbmc3k,reference = 'pbmcref')
#Upload pbmc3k.h5Seurat through Shiny App (https://app.azimuth.hubmapconsortium.org/app/human-pbmc) , map cells to reference, then download the azimuth predictions.
azimuth_pred_App <- read.delim('azimuth_pred.tsv', row.names = 1)
#make sure we have the same cells
all.equal(rownames(azimuth_pred_App),rownames(pbmc3k@meta.data))
#plot
plot(pbmc3k@meta.data$mapping.score,azimuth_pred_App$mapping.score)
Hi,
Thanks for the great tool. I have run different scRNA datasets (human-lung-v2) using the App and then re-running it locally using the Analysis script template supplied in the Download section of the App.
I noticed that with the same dataset (reference data obtained through SeuratData), and with the same QC filter settings, I obtain different results (i.e. mapping.scores or predicted.ann_level_N.scores run through the App and locally and still highly correlated, but definitely different).
Anyone has tested/experienced this ? Any reason why results would be different and which one to trust?
I am using
Seurat v4.3.0
&Azimuth v0.4.6.
Thanks a lot.