CellOracle KO simulation produced results opposite to real KO results

[Sustella]

Thanks so much for the wonderful package for studying cell differentiation through regulator gene perturbation simulation.

From the literature, Isl1 plays a very important role in differentiating a progenitor cell A into a descendant cell B, i.e. when Isl1 is knocked out, cell A stops differentiating to cell B. Isl1 expression value decreases when differentiating to the descendant cell. When Isl1 KO is simulated in CellOracle, cell identity still shifts from A to B; if Isl1 expression value was set to 5, cell identity shifted from B to A. The results are different from real KO results.

I would like to confirm with you what I did is correct.

1. The input data I have is after applying the normalizedata() function from Seurat. Before using CellOracle, I applied math.expm1() to revert the effect of log1p(). Since I don't have the original raw data, would this work?
2. Since I have cell type annotation, I do Isl1 KO simulation for all the cell populations of the progenitor cell A and descendant cell B, instead of sub-regions of cell A and cell B from clustering. Is it OK?
3. From the results of the CellOracle paper, it seems that all "differentiation inhibited" KO results occur in the input data where regulator expression value increases when differentiating to a descendant cell. In my data, it is the opposite.

Appreciate your feedback.