About Maker gene score and ARI/NMI #5
Description
We've generated synthetic data using the CausCell, but I've noticed it's much denser and less sparse than the original real-world data (MERFISH_Brain). When we proceed to evaluate clustering performance using metrics like ARI and NMI, should we apply the same sparsification or thresholding process to the synthetic data as we did to the real data?
I am uncertain whether the observed ARI/NMI and marker gene scoring performance is a normal phenomenon or an indication of a problem.
Factor_Age: ACC:0.6111796299980108 Precision:0.46706897090878147 Recall:0.6111796299980108 F1:0.5281060326597228
Factor_Domain: ACC:0.5559578277302566 Precision:0.454913594815419 Recall:0.5559578277302566 F1:0.49566973287684635
Factor_Celltype: ACC:0.9172468669186393 Precision:0.8535435465911773 Recall:0.9172468669186393 F1:0.8832583196156427
pcc:0.9096 mse:0.2521
ARI: 0.5482 NMI: 0.6550 （k=8)
Maker gene score:0.5805. (top50)