You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thank you very much for implementing CellCharter!
As I discussed in the past, @LLehner had to implement a simplified version of CellCharter to avoid adding new dependences.
The most important one is replacing scVI's dimensionality reduction with PCA.
In my experience developing CellCharter I noticed that using PCA can considerably decrease the quality of the niches, especially in spatial transcriptomics datasets, and is not as scalable for large datasets.
For this reason I would like to propose to change the name of the flavor parameter in the calculate_niche function from "cellcharter" to "cellcharter_simple".
This is to discourage people from using the squidpy's implementation for benchmarking CellCharter against other methods.
In addition, assuming that people using squidpy's implementation will have less familiarity with CellCharter, I would propose using a more good-enoughdistance=3 as a default value instead of the current distance=1 (the docs currently say that it's required but from the code it looks like 1 is set by default).
A PR is on the way!
Let me know what you think :)
The text was updated successfully, but these errors were encountered:
Hi @marcovarrone yes we still wanted to change that. The CellCharter implementation is not final, the name change will definitely be added! We likely also change the GMM implementation. Btw have you ever tested a spatial version of PCA? I thought this might be interesting to use as a lite version of CellCharter, instead of using classic PCA. And thank you for the note on default params!
Thank you very much for implementing CellCharter!
As I discussed in the past, @LLehner had to implement a simplified version of CellCharter to avoid adding new dependences.
The most important one is replacing scVI's dimensionality reduction with PCA.
In my experience developing CellCharter I noticed that using PCA can considerably decrease the quality of the niches, especially in spatial transcriptomics datasets, and is not as scalable for large datasets.
For this reason I would like to propose to change the name of the
flavor
parameter in thecalculate_niche
function from"cellcharter"
to"cellcharter_simple"
.This is to discourage people from using the squidpy's implementation for benchmarking CellCharter against other methods.
In addition, assuming that people using squidpy's implementation will have less familiarity with CellCharter, I would propose using a more good-enough
distance=3
as a default value instead of the currentdistance=1
(the docs currently say that it's required but from the code it looks like 1 is set by default).A PR is on the way!
Let me know what you think :)
The text was updated successfully, but these errors were encountered: