Open WallnerMarkus opened 11 months ago
Hello @WallnerMarkus , There are issues caused by dependencies implementing some of the jDR method's, some of their packages have been deprecated or haven't been further updated for compatibility. I think this is the case for MOFA, MSFA and iCluster. If you are interested in running the benchmark for some other jDR methods while this is fixed, you could manually remove these dependencies from the code and generate another environment without them, something like:
conda create --name momix python r-base r-essentials scikit-learn -c conda-forge
then with this environment activated manually install the R dependencies:
if (!require("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("Biobase")
BiocManager::install("omicade4")
BiocManager::install("iClusterPlus")
install.packages(c("RGCCA", "r.jive", "IntNMF", "GPArotation", "tensorBSS", "InterSIM"))
and for python install the scikit-fusion package from https://github.com/mims-harvard/scikit-fusion
Hello,
I add to the solution posted by @dlsoltero that I was to recreate a fully functioning momix environment through manual installation of all the dependencies in a system running Ubuntu Linux v20.04.
I added my script in a pull request (here) to make it available directly in the gitHub repoasitory.
In case you get any errors during installation on a Windows machine, you can run it on a WSL shell (https://ubuntu.com/wsl).
I also add that to keep the package running I had to use later versions of some packages (both R and Python) so carefully rerun the notebooks using the test data to ensure that everything is working as expected before using them on other datasets.
Best,
Daniele
Thanks for the answers. I was able to setup the environment the way @wariobrega proposed here. I just skipped MOFA since it it seems to use a very deprecated sklearn version. Therefore i removed that part from the runfactorization.R The simulated data notebook seems to run without a problem. For the Cancer notebook however i get a lot of warnings. Benchmark_cancer_warnings.txt
Are you able to run the notebooks without a problem to reproduce the results?
Best Markus
Hello @WallnerMarkus,
As far as I can see, these warnings should not produce any significant changes in the output of the notebook.
They are related to the nperm
parameter in the fgsea
package. As reported here, that parameter is deprecated and you can remove it.
Hope this helps,
Daniele
Hello, we want to use your environment and the notebooks to test the performance of other jDR Methods. Unfortunately we were not able to get the environment created with the current miniconda version and the command you provided in the README. Is there possibly a conflict with the python 3.11.4 of the current miniconda version? On my Windows PC i got the following result: