[gmx-developers] Re: Adding an artificial energy term for structure refinement

[alex.bjorling]

Anton Feenstra-3 wrote
> 
> On 16/08/12 10:22, alex.bjorling wrote:
>> Forgot to mention that the artificial energy term would of course be
>> based on
>> some running average of predicted scattering, not the predicted
>> scattering
>> of individual frames. My question remains, though, of whether GROMACS
>> needs
>> explicit forces.
> 
> The short answer is: yes.
> 
> But you are not clear on what you want to achieve.
> 
> Do you want to generate an ensemble of conformations that are compatible 
> with your experimental data?
> 
> In the first case, perhaps a Monte-Carlo approach is more suitable.
> 
> In the second case, you don't need those artificial forces. Just run
> your simulation and trust that Newtonian mechanics gets 'the
> time-averaging right' as well as 'the fitted structures make some
> mechanical sense'. ;-)
> 
> 

Sorry for being unclear.

The aim is solution-structure refinement from experimental data, so
basically your first case. Certainly, Monte Carlo simulations might be a
better strategy, and I have yet to decide what to do here. But sticking with
MD for the moment, I think there are two issues.

a) Force fields and data are not perfect, and they won't coincide in
general. I was thinking of a weighted combination of both. This has been
done for NMR data, but with closed-form force expressions...
b) As I mentioned, one reason for using MD as a refinement tool is to get
some physically sane and averaged representation of unstable intermediates.
The idea is that the artificial energy term would help to choose roughly
which such intermediate to simulate.

Having typed this, I now see that a Monte Carlo approach would be much
wiser, as that would satisfy both these requirements.
Thanks for your time!
Cheers,
Alex



--
View this message in context: http://gromacs.5086.n6.nabble.com/Adding-an-artificial-energy-term-for-structure-refinement-tp5000240p5000246.html
Sent from the GROMACS Developers Forum mailing list archive at Nabble.com.