Split out object models into separate package?

[jchodera]

It would be useful to have a lightweight package for easily reading, interrogating, manipulating, and doing analysis of the JSON analysis files we generate from fah-xchem. To do this, we currently need to install the whole fah-xchem package (including several fragile pip+git installs), which is not friendly for any downstream users.

We should consider splitting the object models into a second package, perhaps harmonizing these with perses and the OpenFE effort.

Use cases might include:

• Browsing the datasets for use in generating new compound designs
• Analyzing accuracy of free energy calculations and interrogating failure modes
• Training machine learning models to predict free energies or difficulties of transformations
• Experimenting with better network construction or network free energy analysis methods

[dotsdl]

We could have fah_xchem the library and fah_xchem_schema for the pure-schema structures, but keep them side-by-side in this repo. Do you see any problem with this proposal?

[jchodera]

Main take-away lessons are:

• We don't need the entire object model to always live here
• Much of the schema / object model is actually general, with xchem applications having specific interpretations of various fields (e.g. xchem_project is the specific project name, but all projects will have names; xchem_fragment_id is just a unique ID for the reference X-ray structure used)
• We can refactor within this repo to generalize the object model and then eventually migrate into general FAH alchemy infrastructure as appropriate
• Nothing is urgent