Enquiry on behavior of tl.recover_dynamics and tl.velocity and whether or not to subset datasets in the dynamical model

[JesseRop]

Dear scvelo/cellrank developers,

Many thanks for developing and maintaining these wonderful tools that allow the elucidation of interesting biological truths!

I'm using scvelo for RNA velocity analysis on Plasmodium and I have a couple of questions

1. From the documentation, it seems that the steady-state model is used to estimate 'velocity_genes' while running tl.recover_dynamics . However, it is not possible to change the parameters min_r2, perc and others that can be adjusted for the steady-state model. Is this because the 'velocity_genes' candidates are just an approximation and adjusting these parameters won't affect the inference of the dynamical model that much?
2. It is also not clear from the documentation whether the arguments needed for estimation of the steady-state above (min_r2, perc ) which can be adjusted for the tl.velocity function will contribute anything when the 'dynamical' mode is used. In other words, will specifying these arguments change the output of scv.tl.velocity(obj, mode='dynamical')?
3. Where is the steady-state slope stored after running tl.recover_dynamics . Are these the 'fit_steady_u' and 'fit_steady_s'. If so, kindly help me understand why there are 2 for each gene instead of only one.
4. I'm only interested in stages/clusters where cells are transitioning to different fates and finding genes involved in the decision to choose one or the other. I'm currently subsetting my dataset to retain only those clusters involved in fate decisions and performing RNA velocity analysis using the dynamical model. I'm wondering whether it may be better to perform RNA velocity analysis on all stages/clusters and then use cellrank to find lineage drivers only in the clusters involved in fate decision.
5. Proposals from this recently published review, https://doi.org/10.1101/2022.02.12.480214 , seems to suggest some changes to standard RNA velocity analysis workflow. Are there any changes to the standard pipeline in the scvelo tutorials that may be helpful given this new information.

Many thanks!
Jesse