2022-09-22 Protein FF meeting note

Protein parameter fits

@Chapin Cavender

• CC will link slides here

• Slide 1

  • CC – Workers are hanging in queue - Maybe UCI would be faster?

  • DM – Running on CPUs? How many are you able to utilize?

  • CC – At UCSD I’m only able to get 5-10 jobs (cores?) at a time. I heard that, when PB runs, he’s getting on the order of 100.

  • DM – We should have room at UCI to run hundreds/thousands. I’ll get to work on getting you UCI cluster access.

  • JW – Is there a higher priority queue that you can submit on?

  • CC – There are different tiers. The head ForceBalance job is run on a paid queue, and the workers are dispatched to the free pre-emptible queue. I’ve been experimenting with promoting the worker jobs to the paid queue and that does help it run faster.

• Slide 3

  • MG – In some sense, the changing the sign of k is equivalent to changing phase by 180, right? I’ll need to think about that.

  • CC – That’s right, but remember that these are also corrections to the 1-4 interactions. But I generally agree that we should look into why these signs are different.

  • JW – Also worth remembering a few weeks ago when CC showed that AMBER library charges seemed overpolarized on backbone donor/acceptor groups.

  •  

• Slide 7

  • DM – Agree that we should do periodicity 1-4

  • DM – Also agree that we should go with the least parsimonious model

    • JW – Agree

  •  

• JW – I said that I’d share dataset standards from protein ligand benchmarks/F@H interface group but it seems to still be in flux. So no guidance there yet. So use your best judgement on dataset compilation and let me know if you’d like advice.

•