fix: AIZynth

[finlayiainmaclean]

This PR:

Fixes AIZynthFinder, by replacing the separate mamba env, and updating the config format.

[finlayiainmaclean]

We're now calling aizynth directly from the molscore env.

[MorganCThomas]

I was thinking of moving aizynthfinder to a flask server to remove MolScore dependencies. I.e., to keep it in a dedicated conda/mamba environment but have it running in flask to receive molecules and return scores.

I also wanted to move away from calling the aizynthfinder CLI line each time and run from the python API if possible, which will avoid model loading/unloading overhead.

[finlayiainmaclean]

I was thinking of moving aizynthfinder to a flask server to remove MolScore dependencies. I.e., to keep it in a dedicated conda/mamba environment but have it running in flask to receive molecules and return scores.

How would you like me to proceed? Shall I make a thin flask server and put it in molscore/servers, and change AIZynth scoring function to inherit from BaseServerSF (akin to how molscore/scoring_functions/legacy_qsar.py works)?

I also wanted to move away from calling the aizynthfinder CLI line each time and run from the python API if possible, which will avoid model loading/unloading overhead.

This PR does exactly this, loading self.finder = self._setup_finder() once per class instance.

[MorganCThomas]

How would you like me to proceed? Shall I make a thin flask server and put it in molscore/servers, and change AIZynth scoring function to inherit from BaseServerSF (akin to how molscore/scoring_functions/legacy_qsar.py works)?

Yes exactly this!

This PR does exactly this, loading self.finder = self._setup_finder() once per class instance.

Perfect! I hadn't seen the code proposed code changes yet, looks good.