Just a follow up for people interested in RISM.
While RISM is sensitive to conformational changes, on closer inspection the issue here was of numerical convergence. I recommend the following to ensure accurate results.
1) The requested tolerance roughly gives the relative numerical precision of the calculation. E.g., a tolerance of 1e-6 with produce excess chemical potentials with relative numerical error on the order of 1e-6 (5-6 digits of precision).
2) Try a sample calculation with double the solvent buffer length you are planning to use. The results from the two grid sizes should agree within the numerical error expected from 1).
3) For aqueous salt solutions, the grid length of the 1D-RISM calculation is extremely important. A good indicator of an appropriate grid length is the DELHV0 field in the Xvv file. Doubling the grid size should not significantly change DELHV0 (5-6 digits of agreement should be expected). Also, DELHV0 values for oppositely charged ions will agree to several digits. This ensures charge conservation in 3D-RISM, without which satisfying step 2) is not possible.
Hopefully this is helpful,
Tyler
> On Feb 1, 2016, at 5:16 AM, Josh Berryman <the.real.josh.berryman.gmail.com> wrote:
>
> Replying to list for the record (I sent structs and inputs to Tyler and
> Dave so they could have a look, some helpful discussion followed).
>
> 1) 3D RISM energies can be very sensitive to conformation, especially at
> high concentrations of multiple species in the solvent, and especially with
> cavities or odd-shaped surfaces in the solute. This is not entirely a bad
> thing and might be at least somewhat physical, but is a pitfall when doing
> single-point energy evaluation of structures generated with other means
> such as GBSA.
>
> 2) Sampling with MD-RISM is manageable and avoids massive jumps in energy:
> the potential surface described is at least somewhat smooth in the
> valleys. If minimisations or singlepoint energy evaluations are behaving
> suspiciously, my takehome from this is to just pick the lowest-energy snap
> and do MD from it as a starting point.
>
> I got (apparently) adequate results using this input file, simplified
> somewhat from the one in the manual as I wanted to avoid multiplying my
> problems with too much novelty:
>
> comment here
>
> &cntrl
>
> ntx=1, ntpr=10, ntwx=10, ntwr=10, nstlim=2000, dt=0.001,
>
> ntt=3, tempi=300, temp0=300, gamma_ln=10., ig=3233, !Langevin MD
>
> ntb=0,
> !Non-periodic
>
> cut=999., !solute-solute interactions
>
> irism=1,
>
> /
>
> &rism
>
> rismnrespa=2,
>
> fcestride=0,fcecrd=2,
>
> /
>
> "RISM! You have it or you don't, that's a fallacy" .... Gorrillaz, Clint
> Eastwood.
>
> Josh
>
>
>
>
>
>
>
>
> On 27 January 2016 at 11:05, Josh Berryman <the.real.josh.berryman.gmail.com
>> wrote:
>
>> Hello RISM experts,
>>
>> I am running 3DRISM calculations on snaps taken from an MD run, to try and
>> estimate solvation energy. The snaps were slightly altered after the MD
>> (extra proton, appropriate change to .top file, followed by minimise in
>> Generalised Born with heavy atoms restrained).
>>
>> I'm using an .xvv for SPC water with 1M of Cl- ions present. The peptide
>> has net positive charge.
>>
>> GB energy is well-behaved, but I get wildly different RISM energies (100s
>> kcal/mol) for different groups of very similar-looking snaps, this is
>> traceable to specific terms in the output:
>>
>> --solvent PE
>> --Partial Molar Volume
>> --dcf Integral
>>
>> Kirkwood-Buff is not affected, it is fairly smooth.
>>
>> What could cause such big changes in the PMV terms? I don't see anything
>> special by plotting out the volumetric info although I am not an expert.
>>
>> I'm cautious about just going ahead and do a RISM minimise because the
>> outliers are *down* in energy from the main bulk of the snaps (and its
>> expensive, of course).
>>
>> Josh
>>
>> **********************************************
>>
>> Input file is pretty standard afaik:
>> &cntrl
>>
>> ntx=1, ntpr=1, ntwx=1,
>>
>> imin=5, maxcyc=1, !Single-point energy calculation
>>
>> ntb=0, !Non-periodic
>>
>> cut=9999., !solute-solute interactions
>>
>> irism=1,
>>
>> /
>>
>> &rism
>>
>> tolerance=1e-4, !1e-4 Saves some time compared to 1e-5
>>
>> apply_rism_force=0, !Saves some time.
>>
>> npropagate=1, !Saves some time and 4*8*Nbox bytes
>>
>> ntwrism=1,
>>
>> write_thermo=1,
>>
>> /
>>
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Feb 03 2016 - 12:00:04 PST
