Hi David,
I've responded to your questions inline:
On Mon, Sep 14, 2015 at 3:31 AM, David Poole <thepoole.ucdavis.edu> wrote:
> Hello Everybody,
>
> I'd been continuing my effort to produce some specialized heme parameters
> and decided that it might be the simplest to use paramfit and a large
> number of QM derived structures+energies. But after the QM is done, I've
> questions.
>
>
> 1) During the optimization process Gaussian produces a number of structures
> with energies that are sub-optimal configurations, is there any reason why
> these cannot or should not be used for parameter fitting? or other concerns
> in that regard?
>
Paramfit doesn't know anything about the underlying chemistry or dynamics
the parameters are supposed to accurately describe. All it can do is fit an
energy function as best as it can. So, if non-representative conformations
are input, it will do its best to get parameters that describe those
conformations, which can result in poor description of actually relevant
regions of conformation space that are sampled during simulation.
If you think the conformations are sub-optimal, you can weight them so that
they are less important to the fitting algorithm. See this section of the
tutorial for how high-energy / bad conformers can bias a fit and how to
weight them less for better parameters overall:
http://ambermd.org/tutorials/advanced/tutorial23/paramfit_5.html
>
> 2) Is there a way to run paramfit on multiple processors?
>
Paramfit can use OpenMP to parallelize the energy calculation across
available CPU cores. You need to
configure AmberTools with OpenMP support (./configure -openmp) and
recompile. Use the OMP_NUM_THREADS environment variable to configure how
many cores to use, with all being used by default. Look for a line near the
beginning of program output stating OpenMP is being used.
>
> 3) I derived the correction constant K using the method described in the
> tutorial, and then ran it to find parameters by list as well as the default
> (since I want to optimize all parameters), in both circumstances it set the
> initial value of K to zero, this obviously deviates from the output files
> in the tutorial.. so I probably need help with this.
>
That is most likely a bug that I have since fixed. Are you on the latest
version of AmberTools with all available patches?
> 4) I am getting some warning related to dihedral spans lacking data
> coverage, given that Is fitting dehedral parameters to the heme system
> unnecessary or is this not a significant message?
>
Given the bond and angle constraints present on the heme group I wouldn't
worry about sampling with those dihedrals as those bonds are not rotatable.
>
> 5) Any other pointers would be greatly appreciated.
>
> If you are performing the fit on iron heme, be very careful with your
quantum calculations as it can be tricky to accurately characterize the
iron atom's spin state. You will probably have to pick whichever state
predominates in your system of interest, or fit bonded parameters for zinc
heme. Have you seen these parameters for P450 heme?
http://www.ncbi.nlm.nih.gov/pubmed/15812779
Best,
Robin Betz
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Sep 14 2015 - 15:00:04 PDT
