From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Mon Apr 30 2018 - 10:34:58 CDT
On Mon, Apr 30, 2018 at 10:50 AM, Zachary Ulissi <zulissi_at_gmail.com> wrote:
> Hi - currently we have ~250 atoms in the colvar, though considering just
> some portion of each molecule we can push that down. The speed is
> especially apparent right now because we're doing implicit solvent, but it
> will be a smaller fraction when we do explicit and force call times go up.
>
Yes, with explicit solvent, 250 atoms for a collective variable that is
O(N) is not even worth parallelizing. You most likely won't notice the
difference.
> Yes, P is an even integer. We are using this as a softer metric for the
> size of a group of molecules than the outer limit (containing sphere). We
> also define/use this in 1 or 2 dimensions (so radius of gyration in a
> plane, or radius of gyration along a line, using the XY or Z components
> only, respectively). Yes, adding it as an option would be great!
>
P can be added easily: but as for selecting specific components, that's
best done in a separate class so that the checks for X, Y, and Z can be
skipped. Also, the variable would not be rotationally invariant: do you
plan on using rotational fitting?
> Thanks for the tip about PLUMED; I'll take a closer look.
>
> Regarding GPR/NN I was thinking something along these lines:
> https://aip.scitation.org/doi/abs/10.1063/1.5020733
> https://pubs.acs.org/doi/abs/10.1021/ct500438v
> https://aip.scitation.org/doi/pdf/10.1063/1.5018708
>
> GPR feels more reasonable to me than NN for guarantees, bounded nature,
> uncertainty propagation, incorporation of prior information on correlation
> structure, etc.
>
All those papers mention publicly available software implementations, so I
would check those first to minimize effort.
>
> Zack
>
>
> On Mon, Apr 30, 2018 at 10:32 AM Giacomo Fiorin <giacomo.fiorin_at_gmail.com>
> wrote:
>
>> Hi Zack, how large is the number of atoms? Note that the regular radius
>> of gyration code is not parallel either, although it would be definitely
>> faster than a custom function via Lepton.
>>
>> Is the exponent P an even integer? It wouldn't be a big deal to have P
>> as an option, rather than being hard-coded. It may even be possible to
>> calculate the inverse gradients for the corresponding expression, and
>> estimate free-energies through the internal forces (as in straight ABF)
>> rather than the applied ones (as in eABF or metadynamics).
>>
>> FYI, PLUMED supports Lepton expressions, as well. Support for it was
>> added immediately after the corresponding one in Colvars, so it should
>> behave more or less the same way given that Lepton has a simple API.
>>
>> If you can define in more detail the NN or GPR methods that you plan on
>> using, I can recommend whether it's easier to port it into NAMD or switch
>> to another MD engine.
>>
>> Giacomo
>>
>>
>>
>>
>>
>> On Mon, Apr 30, 2018 at 9:57 AM, Zachary Ulissi <zulissi_at_gmail.com>
>> wrote:
>>
>>> Hi - we're currently doing ABF simulations using NAMD/colvars. We're
>>> using something like a generalized radius of gyration of a group of
>>> particles - same definition, but a power P greater than 2:
>>> R_g = [ \sum (R-R_{COM})^P ] ^(1/P)
>>>
>>> 1) We currently have this implemented using the lepton custom function
>>> code in the NAMD colvars module, which works well but is not parallel and
>>> does not scale well with # of particles. Does anyone have a suggestion for
>>> improving this? Hand-coding a TCL custom colvar was definitely slower.
>>>
>>> 2) We're planning to repeat this calculation many times with differences
>>> in structures, so we will have significant prior information about the
>>> shape and smoothness of the final potential. Using something like a
>>> gaussian process/ kriging seems like an obvious way to accelerate the
>>> process (and we can input smoothness ahead of time), but I don't see that
>>> implemented in colvars. The NN code in SSAGES looks interesting but as far
>>> as I know does not support NAMD. PLUMED looks similar but I don't think
>>> there is something like lepton or the generalized radius of gyration above,
>>> so I'm a little hesitant to jump in. Are there other colvar/metadynamics
>>> codes I should be considering?
>>>
>>> Thanks
>>>
>>> Zack
>>>
>>>
>>> --
>>> Zachary W. Ulissi
>>> Assistant Professor
>>> Department of Chemical Engineering
>>> Carnegie Mellon University
>>> Pittsburgh PA 15213
>>> (412) 268-9517
>>> https://ulissigroup.cheme.cmu.edu
>>>
>>
>>
>>
>> --
>> Giacomo Fiorin
>> Associate Professor of Research, Temple University, Philadelphia, PA
>> Contractor, National Institutes of Health, Bethesda, MD
>> http://goo.gl/Q3TBQU
>> https://github.com/giacomofiorin
>>
> --
> Zachary W. Ulissi
> Assistant Professor
> Department of Chemical Engineering
> Carnegie Mellon University
> Pittsburgh PA 15213
> (412) 268-9517
> https://ulissigroup.cheme.cmu.edu
>
-- Giacomo Fiorin Associate Professor of Research, Temple University, Philadelphia, PA Contractor, National Institutes of Health, Bethesda, MD http://goo.gl/Q3TBQU https://github.com/giacomofiorin
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:21:03 CST