Logging simulation metadata on F@H

[JenkeScheen]

In broad terms, what are you trying to do? On F@H, log a collection of information on (finished) free energy calculations, related to https://github.com/openforcefield/fah-alchemy/issues/2. The idea is that given enough gathered information, knowledge could be distilled to create a well-informed atom mapping scorer that can extend the lomap-based and perses-based scorers in OFE (similar to this work). To this end, the following metadata have been outlined in previous discussions:

• Transformation SMIRKS
• Simulation length/ sampling time (should be coupled to FE protocol because e.g. sampling times in REPEX and NEQ are fundamentally different)
• Rate of convergence
• Error estimates (ideally across replicates)
• FE protocol (with version tag?)
• Force fields
• Protein environment information (if not whole, then ligand-protein interactions?)
• Details on solvent

--> Action point: please add to these.

How do you believe using this project would help you to do this? We would need a large volume of data to be able to train such a model - FE calcs on F@H would be very suitable for this reason. Because FE calculation methods change over time, continuous feedback of simulation metadata would be extremely valuable when trying to evolve such a model as well.

What problems do you anticipate with using this project to achieve the above? Presumably not all edges being run on F@H are intended to be shared (publicly). We would only want to collect non-descriptive data (hence SMIRKS rather than SMARTS), but perhaps an opt-out option would be appropriate as well.

[richardjgowers]

I like the above list, "time" ought to mean both simulated time (ns) and number of decorrelations/fluctuations, but I think maybe that's what you mean by sampling time?

re: SMIRKS, it might also be nice to have SMILES of each end (for ligands), possible needing to be optional