salt concentration

[francesco oteri]

Dear OpenMM,
when I run implicit solvent MD, is there any way to account for salt concentration?

[John Chodera]

Hi Francesco,

At the moment, there is no salt-dependent term in the implicit solvation models supported by OpenMM.

The Amber package has a modified GB model that can incorporate monovalent counterion screening in a Debye-Hückel manner:

Srinivasan, J.; Trevathan, M.W.; Beroza, P.; Case, D.A. Application of a pairwise gener- alized Born model to proteins and nucleic acids: inclusion of salt effects. Theor. Chem. Acc., 1999, 101, 426–434.
http://dx.doi.org/10.1007/s002140050460

In principle, it should be possible to implement these modifications to the GB function though a CustomGBForce. See the documentation and header files for CustomGBForce for more information.

John

[francesco oteri]

Thanks John.

In your opinion, from the performance point of view, is better hacking the code or using the CustomGBForce?

[John Chodera]

Peter Eastman can answer this question much better than I can, so I will defer to him on this one.

John

[francesco oteri]

I am trying to modify the source code to add the salt effect. I'have extracted from AMBER user manual the way to change the actual implementation. The problem is that I'm unable to locate the source file to modify. The best candidate is Forces.cu but I don't understant what is the use of files kCalculateCDLJObcGbsaForces1.*, kCalculateObcGbsaBornSum.*, kCalculateObcGbsaForces2.*


Any Suggestion?

[John Chodera]

Francesco,

The implementation details are complex and potentially still in flux, and are not yet well-documented for the Cuda platform. I would highly suggest making your modifications as a CustomGBForce for now unless you really know what you're doing. The expressions you provide get automatically compiled into kernels, and should be quite efficient for the OpenCL platform.

John

[Peter Eastman]

Hacking it directly into the CUDA code will give better performance, but using CustomGBForce with the OpenCL platform will be *much* easier to implement, not to mention being easier to maintain. So I suggest you start with that as a way to get something working, and to try it out and see if you're satisfied with the results it gives. If you conclude that you're happy with it and you're willing to put in the work to get the best possible performance, I can give advice about how to proceed then.

Peter

[francesco oteri]

OK..I will try!!!
Thank you all for the help.
I will report the results