-
Notifications
You must be signed in to change notification settings - Fork 50
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Very large (creating entire rings) proposals #53
Comments
+1 for naming this file "gefitini". This will definitely be a drink on Friday's menu. |
Ring creation will definitely need some TLC. We may need to do more than just consider impropers. |
I wish I had done that on purpose...
Yeah, you're probably right about that. |
The big question here: If we add impropers, are there still multiple ways the bonds around the ring could go so that the ring doesn't close properly, or is there only one maximum in the torsion |
The best model system to start to test on is probably |
I think we might benefit from looking into ensemble chain-growth or Rosenbluth-type methods for growing a good proposal configuration. Here's a review of pruned-enriched Rosenbluth methods: http://arxiv.org/abs/1107.1214 The difficulty would be once we start using ensemble-growth and/or pruning-type methods, it may be difficult to compute the Hastings proposal ratio for the Metropolis-Hastings acceptance criteria. Some of these methods track weights or an estimate of the partition function for the generated ensemble, so it is possible this could provide what we need. |
Another possibility is to use parallel Monte Carlo to generate the geometry proposal before relaxing it with NCMC. If we only used the valence potential energy for this part, we that would remove the contribution of valence energy terms from the overall NCMC + RJMC + NCMC log acceptance probability, and still allow us to use NCMC to anneal the other contributions from intermolecular interactions. |
Another idea: currently, there are no potential terms to keep rings in a ring like shake because they are not needed when rings are covalently closed. We could instead introduce a guide field potential that functions solely to prevent rings from being created in a malformed geometry like two of the three examples above. It should really only kick in for the malformed cases, and should still be finite everywhere, but sufficiently large to greatly increase the yield of properly closed rings. It might require some brainstorming and analysis of MCSS pairings to figure out the simplest way to implement this, but it would be very simple from an implementation and algorithmic point of view. |
Thanks for the suggestions--a lot of interesting stuff! Going through the PERM review now, and gathering some thoughts on this... |
Lots and lots of great stuff on the related configurational-bias Monte Carlo (CBMC) here: |
Some slides with an overview of methods here: http://www.acmm.nl/molsim/molsim2011/Presentations/Lecture8.pdf |
From the Towhee CBMC page:
|
This is resolved. |
CBMC stuff may still be useful. Maybe split this off into separate issue? |
Oh, I agree completely that CBMC stuff is still useful. I'm personally very interested in it and would love to see where we could take it. That said, I'd rather not have a very general theory question as an issue that could hang around forever--I'd personally prefer to keep the issues to near-term goals and software problems. I'm not wedded to this view--if you think that's a bad plan and that it'd be best to have these discussions here feel free to reopen. |
Or open a new issue/whatever you think is best. |
This plan is fine with me. |
So, I just tried a very large proposal of transforming erlotinib into gefitinib. A couple things:
For viewing pleasure, the results are available here. The configuration after dynamics looks pretty messed up, and may be on its way to NaNdyland.
The text was updated successfully, but these errors were encountered: