Hello,
I have zebrafish multiome data for a developmental blood system. My data is 6 timepoints ranging from 12hpf to 30hpf for blood and vasculature populations. The data has been integrated in Seurat and ran through multivelo to get latent times.
My goal was to use CellRank to determine fate probabilities of our two earliest progenitor populations captured in these data to see if there is a bias towards one lineage or another. Since this system is highly plastic I think the algorithm is struggling a bit to accurately determine terminal states.
Some additional details:
-
We have a single sample per timepoint so it is possible some biology got smoothed out during integration, however latent time values correlate strongly with biological timepoint.
-
I am using the RealTimeKernel. When I plot out the random walk, the start and end points look reasonable given our cell type annotations.
-
I have tried the GPCCA and CFLARE estimators, but I do not see an eigengap in either, which if my understanding is correct, is expected for a system actively undergoing rapid development.
-
I have tried manually setting initial and terminal states (if they were not called as macro states), but this seems to set the entire cluster. I’m not sure if this is expected or appropriate to then calculate fate probabilities on?
My question: Is this type of analysis doable in CellRank for a system like this? Would overriding the macro state calculation and manually setting initial and terminal states be appropriate if we have some a priori knowledge of the underlying biology?