[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [IMP-dev] Harmonic





On Thu, Jan 8, 2009 at 7:58 PM, Daniel Russel <">> wrote:
Having such a general function in IMP would be very nice (I was using a non-smooth truncated harmonic before). Such a function should not be dependent on the parameter tuning (since the parameters would simply be passes by a user). 

Davide, if you have a version of your code which allows the user to pass in the threshold and cutoff, we can put a version into core for people to look at as a starting point. Otherwise, it would only take me a few minutes to throw something together.

It sounds like we would like something like:

class TruncatedHarmonicLowerBound {
   /** threshold > center, capped_value > .5*k*(center-threshold)^2 */
   TruncatedHarmonicLowerBound(Float center, Float k, Float threshold, Float capped_value)
}
that'd be great. for me that'd help a lot

If we use Frido's smooth function, capped_value could default to k*(threshold-center)^2, i.e. 2x the value at the threshold to give it space to go smoothly to the value. And we could have Upper and symmetric versions too (as that is little more code). 
also fine with me.
for the specific function: i just came up with 2-x^(-2) spontaneously. i'd also be happy with another solution (morse, fermi, ...) if that is advantageous. if anyone has a probabilistic reasoning for a specific function, that'd be nice.

thanks

frido


On Jan 8, 2009, at 3:44 AM, Davide Baú wrote:

It's the exp() function of the standard cmath.
Performance is indeed another thing to be improved.

Davide



On Thu, Jan 8, 2009 at 12:37 PM, Friedrich Foerster <" target="_blank">> wrote:
thanks.

just for info: isn't the exp() a bit slow, or are you using some dodgy macro?

best

frido


On Jan 8, 2009, at 12:05 PM, Davide Baú wrote:

Hi Frido,

these are the methods I implemented so far in a new class Morse.h.

The first one returns a constant score if feature is greater then threshold. At the moment the returned score is the Harmonic potential calculated when feature >= threshold (i.e. when feature is >= threshold, the returned score will always be Harmonic::evaluate(threshold) ).
The second one implements the Morse potential as described by Marc.

These methods are still in a development phase since there are some parameters which need to be better defined, such as the threshold to be applied and the parameters that define the Morse potential.

Regards,
Davide


 // 1. a function that is a Harmonic for x<std and constant for x>std
 /** If the feature is greater than or equal to the threshold, the score is constant.
     \param[in] feature Value of feature being tested.
     \return Score
 */
 virtual Float evaluate(Float feature) const {
   Float k = Harmonic::get_k();
   Float mean = Harmonic::get_mean();
   Float threshold = 1.0 * mean + XYZRDecorator::get_radius();
   if ( feature <= threshold ) {
     return Harmonic::evaluate(feature);
   } else {
     return evaluate_with_derivative(threshold, k).first;
   }
 }


 //  2. a function that is a Harmonic for x<std and asymptotically approaches a constant value for x-> inf:
 /** If the feature is greater than or equal to the threshold, apply the Morse potential.
     \param[in] feature Value of feature being tested.
     \return Score
 */
 virtual Float evaluate(Float feature) const {
   Float mean = Harmonic::get_mean();
   Float k = Harmonic::get_k();
   Float threshold = 1.0 * mean + XYZRDecorator::get_radius();

  // MORSE POTENTIAL PARAMETERS
   Float var = 1.0/k;
   Float alpha = 2.0;
   Float De = 40.0;
   Float beta = 1.0;
   Float swf = 0.0;

   if ( feature <= threshold ) {
     return Harmonic::evaluate(feature);
   } else {
     swf = beta * pow( ( 1.0 - exp( - sqrt( alpha/De ) * ( feature - threshold )/2.0 ) ), 2 ); // MORSE POTENTIAL
     return swf;
   }
 }





On Thu, Jan 8, 2009 at 10:20 AM, Marc A. Marti-Renom <" target="_blank">> wrote:
Hi Frido,

Indeed we are also in need of similar stuff. Davide Bau (a postdoc in my lab and also in the IMP-Dev list) has already implemented a sort of a MORSE potential in IMP restraints, which is not what you ask but may be similar. I implemented many many many years ago something similar in CHARMM (http://sgu.bioinfo.cipf.es/martirenom/Thesis/pdf/Thesis.pdf, Chapter 4 page 86).

Unfortunately, we are NOT yet ready to commit the changes in the code to the SVN since we are in the middle of production phase (ie, making models) thus would prefer not to update IMP. But I am sure we could do soon.

Davide, any comments? :-)

m

-------
Marc A. Marti-Renom, Head of the Structural Genomics Unit
Bioinformatics & Genomics Department, Prince Felipe Research Center
Avda. Autopista del Saler, 16 , 46012 Valencia, Spain
Tel +34 96 328 96 80  (ext. 1015) Fax +34 96 328 97 01
email " target="_blank"> web http://sgu.bioinfo.cipf.es

*** Consider the environment. Print responsibly. ***

On Jan 8, 2009, at 10:00 AM, Friedrich Foerster wrote:

hi all,

when dealing with experimental restraints i often feel that harmonic restraints are somewhat unsatisfactory:
experimental data can be wrong and thus there is not necessarily a solution that fulfills all restraints (score=0). in those cases, a model with a lower score is not necessarily any more correct than a higher scoring one. thus, i think it'd be good to have an additional scoring function that levels off if the feature exceeds a certain value. for example, i'd suggest the following two functions alternatively to Harmonic:
1. a function that is a Harmonic for x<std and constant for x>std
2. a function that is a Harmonic for x<std and asymptotically approaches a constant value for x-> inf:
  e.g.:  f=x^2 for x<1 and f=2-x^-2 for x>1 (would even be continuous in 1st derivative)

is anyone else also interested in something like that?
any better suggestions?
is any IMP core expert interested and willing to code that (including analogous functions for Upper and Lower Harmonic)? probably i'd be one of the slowest and clumsiest persons in doing that...

thanks

frido


--

Friedrich Foerster
Max-Planck Institut fuer Biochemie
Am Klopferspitz 18
D-82152 Martinsried

Tel: +49 89 8578 2651
Fax: +49 89 8578 2641

" target="_blank">

www.tomotronic.org






_______________________________________________
IMP-dev mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/imp-dev



_______________________________________________
IMP-dev mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/imp-dev


_______________________________________________
IMP-dev mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/imp-dev

--

Friedrich Foerster
Max-Planck Institut fuer Biochemie
Am Klopferspitz 18
D-82152 Martinsried

Tel: +49 89 8578 2651
Fax: +49 89 8578 2641

" target="_blank">

www.tomotronic.org






_______________________________________________
IMP-dev mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/imp-dev


___________
Carpe diem.

Chi non ride non è una persona seria.
Who does not laugh is not a serious person.

"Di solito non sono un uomo religioso, ma se tu sei lassu', salvami, Superman!"
> Homer Simpson
_______________________________________________
IMP-dev mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/imp-dev


_______________________________________________
IMP-dev mailing list
">
https://salilab.org/mailman/listinfo/imp-dev