Various package issues with bioconda r-archr 1.0.3 release

[Pirhhot]

Hello,

I am just writing this post for anyone that would try to use the bioconda release version 1.0.3 of r-archr because you are in for a ride.
Firstly, MACS2 is not compatible with the dependencies of the package (unlucky). As such, my solution was to create a second conda environment and point to the path of MACS2 in the second environment from my R-ArchR environment.
Secondly, by default, bioconda installs r-harmony version 0.1 (as it is the only one available in this channel). However, r-harmony 0.1 used with rcpp 1.0.10-1.0.14 (the only ones compatible with R-4.3.3 necessary for r-archr 1.0.3) will result in Error: element-wise pow(): incompatible matrix dimensions:.
The solution is to manually install a more recent version of r-harmony (available on the conda-forge channel !) with conda install -c conda-forge r-harmony.
Lastly, I had issues with subthreading while using createArrows but adding addArchRLocking(locking = TRUE) seemed to solve the problem.
I have not yet used all of the ArchR's features, importantly I did not try Monocle or chromVAR yet so I will probably come back to this post in case I stumble upon similar issues.
I think this is a good place to share experience regarding the usage of this bioconda release, I would be very happy to see if some people have found other solutions to my issues or if you had other problems.

[immanuelazn]

Hi @pirhot, I can't address the bioconda stuff but maybe @mfansler can.

There are some file locking issues within hdf5 that only occur in specific conditions, that the original authors were unable to figure out. I am currently building the spiritual successor to ArchR and therefore do not have the time to directly fix the threading issues. The file locking with addArchRLocking(locking = TRUE) was the correct move to make.

I would suggest trying out using some of the docker images and renvs on our README, or even installing through pak if you would like a quick solution. From my last check, all functions should be working with them

[mfansler]

@Pirhhot thanks for sharing your experience - it is always helpful to have feedback! I can try to have a look next week.

In the meantime, I can comment for this part:

Secondly, by default, bioconda installs r-harmony version 0.1 (as it is the only one available in this channel). However, r-harmony 0.1 used with rcpp 1.0.10-1.0.14 (the only ones compatible with R-4.3.3 necessary for r-archr 1.0.3) will result in Error: element-wise pow(): incompatible matrix dimensions:.
The solution is to manually install a more recent version of r-harmony (available on the conda-forge channel !) with conda install -c conda-forge r-harmony.

It is important to note that Bioconda does not have a standalone channel. As such, it has very specific directions for using bioconda channel that if not followed can lead to the above issue. Namely, one must always have conda-forge at higher priority than bioconda when solving environments with conda. If you are doing ad hoc installs into an existing environment, and you don't already have channels configured as the documentation recommends, then the command should be in the form:

conda install -c conda-forge -c bioconda r-archr

If that is followed, then the issue you report should not be encountered.

[Pirhhot]

That is VERY helpful, thank you very much !

[mfansler]

Just swinging back from the Bioconda discussion, the issue with incompatible macs2 was really about installing iteratively (not recommended) versus defining all packages at the outset. That had locked an incompatible Python in the environment and Conda by default will not change a Python version in-place without being told.

Things work as expect if macs2 is declared at creation:

conda create -n foo -c conda-forge -c bioconda r-archr macs2