pip install ScreenPro2 installs no runtime dependencies; environment.yml pins also fail to resolve #112

Description

Installing ScreenPro2 from PyPI as the README describes (pip install ScreenPro2) produces a package that can't be imported. pyproject.toml declares no runtime dependencies other than Python >=3.9:

Steps to reproduce

conda create -n sp2 python
conda activate sp2
pip install ScreenPro2
screenpro --help

Observed

ModuleNotFoundError: No module named 'pandas'

Related problems with environment.yml

Installing the dependencies listed in environment.yml doesn't fully solve it either:

1. Unbounded Python version. With no Python version specified, conda picks the newest Python (3.14 here). biobear has no wheels for it, and the pandas<2.0 / matplotlib<3.7 pins can't be satisfied. The range python>=3.9,<=3.12 might also be too loose, since pandas 1.5.3 has no official build for Python 3.12. Python 3.11 works.
2. Unpinned pydeseq2 conflicts with pandas<2.0. The current pydeseq2 (0.5.4) requires pandas>=2.2, so pip install pydeseq2 silently upgrades pandas 1.5.3 → 3.0.6, numpy 1.26 → 2.4.6 and matplotlib 3.6 → 3.11. Pinning pydeseq2<0.5 (0.4.12) resolves it.
3. libstdc++ ABI clash from mixing conda and pip. Because the yml installs some packages via conda and others via pip, the pip wheels (pyarrow/biobear/polars) load the system libstdc++ first, and conda-forge scipy then fails:

   ImportError: /lib/x86_64-linux-gnu/libstdc++.so.6: version `CXXABI_1.3.15' not found
   (required by .../scipy/spatial/_distance_pybind.cpython-311-x86_64-linux-gnu.so)

Workaround: conda env config vars set LD_PRELOAD=$CONDA_PREFIX/lib/libstdc++.so.6

Working configuration

ScreenPro2 0.7.0 runs (screenpro --help works) with:

• Python 3.11 (miniforge, conda-forge + bioconda)
• pandas 1.5.3, numpy 1.26.4, matplotlib 3.6.3, pydeseq2 0.4.12
• biobear 0.23.7, pyarrow 25.0.1, anndata 0.11.4, scipy 1.17.1
• LD_PRELOAD workaround above

Suggested fixes

• Declare runtime dependencies (with the correct version bounds) in [tool.poetry.dependencies], so pip install ScreenPro2 works as the README says.
• In environment.yml, pin python=3.11 (or >=3.9,<3.12) and pydeseq2<0.5, or update the code for pandas ≥2.

Environment

• OS: Linux x86_64 (Ubuntu 22.04.5 LTS)
• Conda: miniforge (conda 23.1.0)
• ScreenPro2: 0.7.0 (PyPI)