I had done most of my spatial transcriptomic celltype mapping reverse enginerring celltrek to get celltypes that map to a visium spot. After the paper submission reviewer wants us to compare our predictions with that of cell2location. Unfortunately the 2 approaches are not directly comparable. E.g I have raw counts of celltypes per spot for celltrek that I can convert to proportions (as I used n cells per spot argument in celltrek to be 8) but I am not sure how I can compare that to cell2location which gives an abundance estimate to each celltype? I originally intended to do presence/absence analysis based on immune marker genes expressed (binary yes/no) for T-cells, B-cells and myeloid cells etc and then compare that with celltrek output and cell2location output to calculate an accuracy and F1 score etc. But the bottleneck is setting thresholds for very small abundance values for cell2location. How would the good folks of scverse approach this issue? Thank you!