Hi Charles, My concern is this :
The force on the conformation whose weight is 0.1 is too low to drive it to the minimum. The possible solution is that increase the force constant by using rdc.setScale or something similar but that would mean the conformation with weight 0.9 would have a large force and am not sure if that would not cause any problems. The multiplication of the weight factor might cause problems in setting the right force constant. While what is being done is mathematically correct, in my experiments the protocol did not work well. We took 2 distinct conformations of a protein and assigned them weights of 0.8, 0.2 while generating the artificial RDC's and NOEs from those. When starting with the global minimum [the original 2 conformations with weights of 0.8, 0.2], the second member moved away from the global minimum. [I hope I was clear in my explanation, please ask if something is not clear]. Any suggestions or insights from you is greatly appreciated. Thanks Santhosh On Tue, Sep 10, 2013 at 8:28 PM, Charles Schwieters <char...@schwieters.org>wrote: > > Hello Santhos-- > > > > > I was able to set up weights during refinement but I have a doubt on how > they > > are implemented in the program. > > > > Suppose I have a 2 member ensemble and their weights are assigned as > > [0.9,0.1]. Now while computing the weighted averages of NOE's and RDC's > the > > individual members contribution is taken care. > > > > But if you used the exact formulation for the force on atoms i.e : > > E_experimental = K_experimental*(ExperimentalData - CalculatedData)**2. > > Where calculatedData = w1 *A1 + w2 *A2. Where w1, w2 are the weights of > > individual members and A1 and A2 are the computed values. > > > > The first derivative of that [force] would have a component of w1 or w2 > in > > the equation. So the conformation with 0.1 weight would face less force > > compared to the other conformation with 0.9. > > > > While what you say is true, I do not see a problem. What precisely is > your concern? > > best regards-- > Charles >
_______________________________________________ Xplor-nih mailing list Xplor-nih@cake.cit.nih.gov http://cake.cit.nih.gov/mailman/listinfo/xplor-nih
