Re: [AMBER] 3D-RISM options

From: Tyler Luchko (Lists) <"Tyler>
Date: Tue, 14 May 2019 15:44:24 -0700

Hi Ryan,

> On May 14, 2019, at 2:45 PM, Ryan Pavlovicz <> wrote:
> Hi,
> I am interested using 3D-RISM for the purpose of predicting hydration sites
> at protein/protein interfaces and would like some help/suggestions on how
> to optimize both runtime and accuracy.

Just so you know, the treecode parameters are new and, AFAIK, there haven’t been any studies on optimizing the speed of water placement calculations. Hydration sites are relatively coarse grained features, so high precision (low tolerance) calculations are probably not necessary.

> I understand that the use of treecode can help speed up the calculation. Is
> the best way to implement this by setting --treeDCF 1 --treeTCF 1
> --treeCoulumb 1? I see a bunch of other treecode-related flags and
> suggestions in the Amber19 manual, but I don't quite understand which
> should be used. Guidance here would be greatly appreciated.

Yes, these flags will turn on treecode summation, just don’t add the ‘1’ for rism3d.snglpnt (this is an error in the manual). In addition, you should set the MAC and Order for each of DCF, TCF, and Coulomb. The values of these depend on the numerical precision you want. Table 7.1 in manual gives some suggestions.

For example, if you are setting the tolerance to 1e-4 then the --treecodeOrderTCF should be -2*(3 + log10(1e-4)) = 2. Likewise, --treecodeOrderDCF should be 4 and --treecodeOrderCoulomb should be 6. Along with the recommended MAC values, this *should* prevent any loss of precision due to using treecode but these are not hard and fast rules. If you are going to do many calculations, it would be worth your time to try some different order values to see if you maintain the precision you want. If you are only going to do a few, you might want to play it safe and add 2 to each parameter; e.g., --treecodeOrderTCF 4.

> I also understand that additional speed up can be accomplished by modifying
> the grid spacing. For the rism3d.snglpnt calculations, would running with
> --grdspc 0.5, 0.5, 0.5 --buffer 7 be reasonable for speedup for water
> placement purposes without sacrificing too much accuracy? What are the
> default values here?

The default grid spacing is 0.5 A, which is probably the most commonly used for water placement. If you are using Placevent, I believe that it always centers waters on a grid site, so the grid spacing will affect the placement of discrete waters. Note that a 0.25 A grid spacing will require 8X computing time and memory.

The default for buffer is 14 A. You can experiment with buffer sizes. If the interface is buried, I think a 7 A buffer should be fine.

> Also, is there any known advantage to using SPCE vs TIP3P when it comes to
> native water recovery? And how sensitive would the water placement be with
> different NaCl concentrations? Is the 5 mM NaCl concentration used in Dan's
> tutorial reasonable?

I don’t know what the effects of ions will be. I would either use pure water or the physiological concentration of ions relevant for your application. Note that including ions will require more calculation time and may require different grid and treecode parameters.

> Finally, as far as output, I only need the O.dx file for placement of water
> positions with Is there a way to silence the H1, Na+, and Cl-
> output to prevent my disk space issues? Relatedly, is the calculation of
> these required if I'm only interested in the O distributions?

There is no way to silence these files and all distributions must be computed. Pure water calculations are at least twice as fast.

Hope this helps,

AMBER mailing list
Received on Tue May 14 2019 - 16:00:02 PDT
Custom Search