search

theislab / scvelo

RNA Velocity generalized through dynamical modeling
https://scvelo.org
BSD 3-Clause "New" or "Revised" License
414 stars 102 forks source link

Velocity changes after update to 0.1.15 #42

Closed JonasMir closed 5 years ago

JonasMir commented 5 years ago

Hi Volker, I just wondered why the RNA velocity flows have changed in my case (not entirely, but significantly) after I have updated scvelo to version 0.1.15 without having done any further changes. Did you change/improve anything that could explain this? Best, Jonas

VolkerBergen commented 5 years ago

Hi Jonas, there have been one or two changes such as setting extreme quantile as default (instead of simple fit). What was your previous version?

JonasMir commented 5 years ago

Ok, I see. My previous version is from October 2018. This is a quite imprecise answer but I don't know how to get this out of pip...

VolkerBergen commented 5 years ago

You can get the version via scv.logging.print_version()

I see that I have made the 'extreme quantile fit' as default in December (being more robust and better at capturing steady states). You can run it on all data (instead of quantiles) via scv.tl.velocity(adata, perc=None).

JonasMir commented 5 years ago

Thanks, this is the answer! Setting perc=None reproduces my old results. Though my system (T cells within acute infection) definitely is not in steady state - which option would you recommend here?

VolkerBergen commented 5 years ago

It's all approximate since a steady-state population is the underlying assumption. Setting perc to extreme quantiles would usually capture the steady states better (or at least be more conservative if no steady state is sampled). You could examine the phase portraits (pl.velocity) for some of your favorite genes and check out the difference when using simple fit vs. extreme quantile fit.

Soon (~March/April) a novel method (capturing the full stochastic dynamics) will be released and published, that is not subject to the steady state assumption anymore.

SamueleSoraggi commented 5 years ago

Looking forward to play around with the method and read the publication!!! Keep the good work up ;)