Python Packaging Issues

[xavierholt]

I'm trying to install Meeko in a Python 3.12 virtual environment on a clean Ubuntu 24.04 system. I've run into a couple problems, all related to package dependencies:

• Running pip install meeko installs Meeko 0.5.0, but this uses from rdkit.six import StringIO and that's incompatible with the latest version of RDKit (2024.3.3).
• Meeko 0.5.1 contains the fix for this, but it won't install via pip because it's listed as incompatible with Python 3.12 in setup.py.
• There are a few other packages that Meeko imports but doesn't list as dependencies, so it'll crash when you run it without these installed.

I was able to get things working by cloning the repo, editing setup.py and installing from source, but it would be nice if things were easier via pip. I think the following would take care of it:

• Remove the Python < 3.12 restriction in setup.py.
• Add pandas, rdkit, and scipy as package dependencies.

It looks like the Python version restriction was added in https://github.com/forlilab/Meeko/commit/03d4168a72cb0d52043852ae388f025e5461e84c to work around ProDy restrictions, so I have no idea how feasible removing it would be... Alternatively, you could pin RDKit to version 2023.9.6 (or earlier), which was the last to include the rdkit.six submodule.

[diogomart]

Thank you for your comment. We have a tough decision to make for meeko v0.6. Our current plan is to require prody (to parse mmcif mostly). But prody doesn't work on 3.12. Removing the <3.12 restriction worked for v0.5 because prody is optional in meeko v0.5 (it's used only for tethered docking). For meeko v0.6 we have to choose between 3.12 or prody, can't have both unfortunately.

[diogomart]

Could Python 3.11 work for you?

[xavierholt]

It looks like Python 3.12 is the only officially supported Python version on Ubuntu 24.04. You can install other versions via unofficial APT repos (a quick Google turned up deadsnakes), but in my case I'm running on AWS, so it's easier to switch to an Ubuntu 22.04 machine than to deal with non-standard packages. This'll be more of a problem for people running on local machines.

Anyway, it sounds like the underlying problem is ProDy, not Meeko. I don't see any tickets regarding Python 3.12 on their GitHub page, so I'll open one and ask.

[diogomart]

Micromamba has been working well for me, where "working well" is in light of the complexity of python dependency networks. For example, this should work out of the box and result in a python 3.11 installation:

micromamba create -n myenv
micromamba activate  myenv
micromamba install meeko

[xavierholt]

Thanks! Using micromamba seems to be easy enough for now. I'm a little nervous 'cause it installed Python 3.11 without me specifying that anywhere, but I can confirm it works on Ubuntu 24.04.

In the meantime, I opened https://github.com/prody/ProDy/issues/1915 asking about Python 3.12 support for ProDy.

[diogomart]

It installed 3.11 because meeko's conda recipe restricts to <3.12. You can have multiple environments with different python versions, or package versions, and pin the python version with micromamba install python=3.12. You can also pip install inside these environments. There's a good chance things will get messed up, eventually, but navigating Python dependencies is nicer with micromamba than without it.