2020-07-01 BespokeFit meeting notes

JH – I’ve been showing JA how qcsubmit works

JH – Most functionality implemented, but no CLI

JH – Will work with Cerutti to submit protein set

• JW will make conda package

JH – Started looking at bespoke fitting pipeline.

• I’m making the optimizer a plugin, which will only support forcebalance for now.

JW – Should we have a baseoptimizer class with a few generic attributes that apply to all optimizers? Then all optimizers that inherit from this could have sensible default values for all the fields that they add.

JW – For the k specification, is there expected to be a correlation between the N in kN and the periodicity?

• JH – Yes. This will output the entire input forcefield, with new torsion k values appended to the end of the hierarchy. The new kN values will have N values corresponding to the periodicity.

• JA – Generating phase from scratch is tricky. Ethene vs. ethane example.

• DM – We shouldn’t worry about fitting phase initially, but infrastructure should keep the door open for that in the future.

• JA – Could autodetect number of minima and maxima to guide selection of periodicity and phase.

JH – So, schema is ready, but haven’t implemented the machinery

JW – Should keep a debug flag threaded into every function. This will not delete temporary directories and will dump QCA database to disk.

Galileo proof of concept

Cerutti coordination

JA – Would we be able to benchmark torsion fits on crystal structure prediction questions?

• JH – Yeah, we could do that.

• DM – Existing CSP methods are largely quantum (or at least different functional forms)

• (General) – We could slot ourselves in to an existing workflow, either as a first-stage crystal form exploration, or as a second stage rescorer

• JW – Is there a standard workflow that we could slot our FFs into? Then we could either show that we do well on an absolute scale, or that we do better than the workflow with a more generic method

• (General) – If we have bandwidth, we could contact the CCDC about getting some structure released to do this prediction against.