2021-06-16 LW/JW

Talk review

• Intro to OpenFF’s architecture? (Molecule, Topology, ForceField)

• Slide 2+3: Maybe give more high-level motivation?

  • Maybe a high-level motivation of “I want to start with polymer building blocks and make a force field”, and maybe a simple diagram of what that might look like.

  • Make a big table of needs with checks and Xs?

• Slide 5: LW – JW do you have more to say here?

• Slide 7: Mobley lab won’t know much about this background (terms: “bespoke”, “SMIRKS”, “LibraryCharges” (maybe an actual librarycharge line from an OFFXML) ) – Maybe add pictures?

• Slide 9: What’s AlogP? Why do you need it?

  • Maybe don’t reference AlogP by name, jsut say “I filtered these by compounds which have experimentally measured properties”

  • SCIENCE: If for validation, how will you isolate effects from partial charges versus those from other FF deficiencies?

    • Two types of “correct”:

      • “Gets the right answer for a particular experiment when paired with a particular FF”

      • “Agrees with RESP”

• Plot: How many molecules in final set?

• JW – Please send me the big outliers on this plot and the code/data of this plot

• Checking how often different atom orderings yield different geometries/charges?

  • JW – I’d recommend doing tehse studies separately – Don’t try to get different geometries by rearranging atoms.

• Slides 3-5 can be removed or distilled down into “future directions”

• Slide 2 could be modified to say “here’s how you NORMALLY use OpenFF with a small molecule, and here’s how I want you to be able to use it for polymer FFs, but the problems are 1) making the polymer parameters and 2) pre-computing charges because the polymer is too big for AM1BCC”

• What additional things would be interesting to show?

  • Fragmentation

    • What do input building blocks/fragments look like?

    • Show a picture of building all possible polymers from building blocks

    • Charge fragment creation styles and capping schemes

  • Combination

    • Talk about averaging over the capping atoms, winding up with non-integer charge in whole polymers, and approaches for dealing with it.

    • Possible open area of research: Should a building block’s parameters depend on its neighbors in some cases?

  • What is the difference/”appropriate”

    • Mention penta-alanine study, conclusion was that the alanine charges were consistent. But this is a limited test so I looked for a more diverse molecule set.

    • Introduce study with fragmenting ChEMBL molecules, say that it’s a proxy for having stuff made from building blocks.

    • Introduce the capping schemes, introduce the joining schemes. .

    • Definitions of “acceptable”

      • Jeff says that anything that produces “AM1BCC” partial charges should agree with other “AM1BCC” methods – within 0.05 to 0.1 is the limit of “acceptable”

      • Mobley paper: 0.01

      • Define the metric that will be used to select the “best” scheme

    • Showing the data

      • Choosing between capping and joining schemes

      • Show some fun plots with lots of data, but that aren’t very interpretable.

      • Show a table and/or cumulative histogram of methods, (x axis is the metric that you mentioned before for selecting the “best”, the best method should be the highest line, if there’s too many lines then color lots of the losers grey)

  • Conclusions

    • Systematic differences (science problems, charge transfer between fragments?)

    • Random differences (technical problems, eg conformer dependence)