Feature Request: Add container-friendly pre-setup utility for PIDGIN dependencies

[naglemi]

Problem

Currently, PIDGIN scoring functions require runtime conda environment creation and model downloads (~11GB), which fails in containerized environments where:

1. Conda environment creation fails due to read-only filesystems or restricted permissions
2. Runtime model downloads are problematic for reproducibility and security
3. First-use delays of several minutes due to environment setup and downloads

Current Behavior

pidgin = PIDGIN(prefix="test", uniprot="P21918")
# Attempts to:
# 1. Run `conda env create -f environment.yml` 
# 2. Download 11GB models from Zenodo
# 3. Launch Flask server

This fails in containers with:

Command 'conda env create -f environment.yml' returned non-zero exit status 1

Proposed Solution

Add a pre-setup utility that container builds can use:

Option 1: CLI utility

molscore-setup pidgin --download-models --create-env

Option 2: Python utility

from molscore.setup import setup_pidgin
setup_pidgin(download_models=True, create_env=True)

Option 3: Container build helper

# In Dockerfile/Singularity definition
RUN python -m molscore.setup pidgin

Implementation Details

The utility should:

1. Create conda environment using existing environment.yml:

   conda env create -f /opt/conda/.../molscore/data/models/PIDGINv5/environment.yml

2. Pre-download models to ~/.pidgin_data/:

   # Trigger model download without server launch
   # Use existing zenodo-client dependency

3. Validate setup:

   # Test that PIDGIN can instantiate without runtime setup

Benefits

• ✅ Container-friendly: No runtime conda operations needed
• ✅ Reproducible: Models baked into container at build time
• ✅ Faster startup: No first-use delays
• ✅ Offline capable: No runtime network dependencies
• ✅ Security: No runtime downloads from external sources

Current Workaround

Users must manually:

1. Create conda environment using the provided environment.yml
2. Somehow trigger model downloads (unclear how)
3. Hope the server-based architecture works in containers

Use Case

This is critical for HPC/cloud deployments using Singularity/Docker containers where PIDGIN scoring is needed for molecular optimization workflows.

Would appreciate guidance on the preferred approach!

[MorganCThomas]

I don't have any conceptual problem with this.

I am not sure I understand why creating a conda environment fails

Command 'conda env create -f environment.yml' returned non-zero exit status 1

But creating it in a utility python script will work... how and where will this utility be run?

conda env create -f /opt/conda/.../molscore/data/models/PIDGINv5/environment.yml

This package hasn't had much consideration of containerized use yet, so it would be a very nice feature.