search

satijalab / sctransform

R package for modeling single cell UMI expression data using regularized negative binomial regression
GNU General Public License v3.0
208 stars 33 forks source link

Problems estimating theta #53

Closed iwillham closed 4 years ago

iwillham commented 4 years ago

Hi Cristoph, Really nice work and paper. I've learned a ton about scRNA-seq data pre-processing by going through your method.

I get the following warning messages after running vst : "In theta.ml(y = y, mu = fit$fitted) : iteration limit reached". In going line by line, I noticed that some of the theta estimates are really large (i.e. 1e15). Does this indicate that these genes are not over-dispersed and instead are more Poisson distributed?

Is this something to be concerned about? Any remedies?

Thanks, ian

ChristophH commented 4 years ago

Hi Ian, Yes, a very high theta (or more accurately, a very small mu^2 / theta) indicates a poisson like model. The warning message is discussed in #25.

iwillham commented 4 years ago

Got it, thank you.