2020-09-23 ANI2x Roche, Janssen Meeting notes

Permanent database or ephemeral?

Josh

• XL: Thinking more about the just getting ANI results for single molecules.

  • JH: Recommend using the Snowflake instance for getting calculations back in a way that’s more like a script.

• Want to be able to have a single callable for torsiondrive evaluation.

  • A single torsion

Demonstration of QCSubmit for ANI2x calculations

Josh

• JH: Use QCSubmit factories to generate Dataset objects with reasonable parameters populated. Validation proceeds on generated datasets before submission to the target QCArchive instance

• Can add molecules to the dataset via molecule objects, .smi files, etc.

• XL: Looks straightforward to run, but what do the dependencies look like?

Aims in industry

Gary

• GT : want to do evaluation of the forcefield on internal molecules

  • can’t have OEToolkit expenses

  • want to be able to rerun scripts for coverage

  • interested in permanent instance for large molecule sweeps

  • take 500 molecules and repeat the analysis done before

  • We want a workflow to set up the archive sub in our molecules and run the analysis and then combine results at the end.

  • also need torsion analysis

• Don’t want to get into a situation of having to install everything at every pharma partner (Janssen’s?)

• DD we are working on the benchmark dash as well which should cover this kind of workflow and add the functions.

• DH : at the moment we only want optimizations torsiondrives will be useful later.

• DD: installing is an issue and so far with workers we have used docker images to help get workers up faster this could also work in this case if the full environment was in docker with an entry point that we can insert the molecues into.

• DH: we are working on a more deployable workflow which I am testing, I have been running torsiondrives locally and that works with the old scripts from Lim.

• DD: lets set up a regular meeting for this time so we can keep a tight feedback loop.

• GT: We want to work out exactly what we want but probably set up the archive and all dependencies then run optimizations would be a great start.