When to set the DefaultAssay to "prediction.score.celltype.l1/2" in MapQuery() workflow?

[denvercal1234GitHub]

Related to #5734, after I performed MapQuery() of my query data onto the ref CITE-seq data (following https://satijalab.org/seurat/articles/multimodal_reference_mapping.html#example-2-mapping-human-bone-marrow-cells-1), I switched to the active assay prediction.score.celltype.l2 to visualize the prediction scores of the transferred labels on my query cells.

Question 1. Now, if I would like to perform DEG to identify genes driving the predicted.celltype.l2 classification in my query cells, do I need to set DefaultAssay() back to SCT before FindMarkers()?

Question 2. Would you mind clarifying when we need to switch to prediction.score.celltype.l1 or predicted_ADT and when to switch to SCT?

Question 3. Does it make sense to sort the resulting DEGs by p_val_adj, then by p_val, then by avg_log2FC, then by pct.1 in order to get the most "important" genes?

Thank you so much!

My current codes after MapQuery():

Idents(Seurat_object_wPredictedcelltypes) <- 'predicted.celltype.l2'

DefaultAssay(Seurat_object_wPredictedcelltypes) <- "SCT"

AllClustersMarkers_markers < FindAllMarkers(Seurat_object_wPredictedcelltypes, only.pos = F, assay="SCT")

# Sorting DEGs 
AllClustersMarkers_sorted <- AllClustersMarkers[order(AllClustersMarkers$p_val_adj, AllClustersMarkers$p_val, -AllClustersMarkers$avg_log2FC, -AllClustersMarkers$pct.1),]