I have questions about the scanpy foldchange computations.
I stumbled across these two issues, which point out two severe issues about the foldchange computation and the
Also, I also experienced, that the foldchanges differ drastically compared to the ones calculated by Seurat or MAST.
Will these issue be addressed in future?
As far as I understood the main difference between the scanpy and seurat or MAST is in the computation of the mean.
- Seurat, MAST using the arithmetic mean.
- Scanpy is using the geometric mean.
The scanpy documentation already points out that tools such as MAST are more reliable in this tutorial.
It would be great to point out somewhere in the documentation that it is not possible to directly compare the logFCs computed with scanpy and those from most R packages.
Please also take a look at the best practice chapter on differential expression analysis.