2020-10-16 Wagner Gokey Check in Meeting notes

What should “Standards” do?

• QCSubmit should be able to load “Standards” versions with a bunch of preset values for how to operate

• A new user should be able to read the “Standards” and submit their own dataset which could be used for fitting a main-line OpenFF force field

• A person should be able to read the “Standards” and do compatible calculations in another engine, like Gaussian

• “Standards” should provide a protocol

• The administrator for the dataset pipeline should look at the Standards and see that completed jobs should have hessians triggered

How can we make standards separable?

• Is a hierarchical structure possible? Like, “use all the settings under the Optimization branch of the hierarchy, but don’t take the entire Hessian hierarchy”

What do we want to make happen?

• Anytime we say “X molecule can be used for a FF fit”, we have a strict checklist for what information it must contain

  • How we make it happen: ???

• Any time someone says “I want to submit a molecule to be used in an FF fit”, we have a strict checklist for how to sub it in such a way to get the information it will need to contain.

  • One path: Make QCSubmit able to apply/enforce this, with human review of squishy things like dataset name/description

  • Another path: Complete human review of submissions, with a checklist that says “make sure it follow these 20 rules”

  • Another path: Some other program/automation to enforce standards

• Anytime someone says “I just want to do optimizations in the normal openFF way”, they are able to take a SUBSET of our standards and apply them

  • We won’t prioritize supporting this, but it’s likely that QCSubmit will have friendly ways to apply the standards and remove some parts of the pipeline

Conforming to the DATASET standards, or the FITTING standards?

• DATASET standards are a MINIMUM that ALL datasets must adhere to, to make sure that we don’t get badly-named or badly-described datasets

  • These will mostly require human review

• FITTING standards are a larger set of standards (a strict superset of DATASET standards) that MUST COMPLETELY be followed by datasets intended for main-line FF fitting.

  • “A checklist to make sure the output will be suitable for ForceBalance”

  • These should be largely machine-validatable

•  

Might be good to make 3 things:

• DATASET standards (human readable document)

• FITTING standards (human readable technical document)

• “Best Practices/tutorial/ADHERENCE” document (practical usage of available tools to make dataset that conforms to standards)

Questions like “do we filter out confs with hydrogen bonds?” are out of scope of this document