I believe the problem lies in the way you have treated the restraints: (1) You should remove stereospecifically assigned NOEs for methylene protons as well as aromatic protons of Phe and Tyr, and treat these as ambiguous NOEs with r-6 sums. If you have made a stereospecific assignment you must automatically have the chi angle and this should then be entered as a torsion angle restraint. The RAMA torsion angle database potential will then take care of everything else. This is actually important because the use of stereospecifically assigned NOEs invariably results in underestimation of upper bounds for distance restraints owing to spin diffusion between the very closely spaced (1.8 A) methylene proton pair.
(2) It is usually best to use rather generous ranges for the NOEs rather than narrow ones. e.g. 1.8-2.5 (2.7 for NH), 1.8-3.3 (3.5 for NH), 1.8-5.0 and 1.8-6.0 corresponding to strong, medium, weak and very weak NOEs. in addition 0.5 A should be added as a correction to the upper bound for every methyl group involved in an NOE (this accounts for the higher apparent intensity of NOEs involving methyl groups). This avoids any systematic underestimation of NOE distances which necessarily results in loss of accuracy. (Note even systematic errors as low as 0.1 A can have a deleterious effect on accuracy). (3) The use of the RAMA term is important because it ensures that regions of conformational space that are sterically not allowed are not sampled unless forced to do so by an NOE(s). If the latter occurs it is almost invariably the case that the NOE is incorrect, unless there is a very good chemical reason for the particular torsion angle to be in an unfavored region of conformational space. (4) To ensure optimal packing and avoid systematic expansion of the structure, the radius of gyration restraint (with a relatively low force constant, e.g. 50) should be applied. Note that if the value of Rgyr is underestimated this has little effect since he hard quartic vdw repulsion term prevents atomic overlap. -- G. Marius Clore, MD, PhD Chief, Protein NMR Section Laboratory of Chemical Physics Bldg 5/Rm B1-30I NIDDK, National Institutes of Health Bethesda, MD 20892-0520 Tel: (301) 496 0782 Fax: (301) 496 0825 E-mail: mariusc at intra.niddk.nih.gov web: http://spin.niddk.nih.gov/clore On 3/18/07 1:12 PM, "Michael Hodsdon" <mehodsdon at yahoo.com> wrote: > > *** > I have made multiple attempts to send this message to the list and I am not > sure what I am doing wrong. So, if it ends up that I have sent a number of > copies of the same message to everyone's inbox, I sincerely apologize. > *** > > Greetings All, > > I have previous experience using CYANA to calculate structures using NOE > restraints. Now, I am trying to include RDCs and am migrating over to > Xplor-NIH. I am having a few difficulties and was hoping to "beg" for some > assistance/advice. > > We are working on a 150 residue protein. We have nearly complete (97%) > assignments and have picked a large number of NOESY correlations from 3D > NOESY-HSQC spectra. Using "NOEASSIGN" (previously known as CANDID) in CYANA, > these NOEs were easily assigned in a fully automated manner resulting in a > nicely converged ensemble of structures (backbone RMSD < 1 Angstrom). > > We next converted the ~3000 independent distance restraints from CYANA to > Xplor-NIH format. This was no easy task given the very different ways that > CYANA and Xplor-NIH handle non-stereospecifically assigned hydrogens. I am > fairly certain that the conversion utility bundled in CYANA 2.1 does NOT > convert the NOEs correctly. I ended up writing my own AWK script to do it. Has > anyone else tried to do this? I am fairly confident that my conversion is > correct as I was able to import the converted restraints into Xplor-NIH and > test them against one of the fitted structures from CYANA (which also had to > be converted!) and got agreement. > > Surprisingly, when I try to use Xplor-NIH to calculate ensembles of structures > using these converted distance restraints, I CANNOT achieve as good > convergence as I could in CYANA. Using a local Beowulf cluster I calculate 100 > structures and keep the best 20-40. I always had 10 - 20 frequently violated > NOEs and a LOT of VDWs violations (20 - 60). I am guessing this is a > consequence of the difference in the potential (pseudo-energy) functions used > for the two methods. For example, in Xplor-NIH I am using this "RAMA" term, > which I don't think there is an equivalent in CYANA. > > I began to suspect that my distance restraints from CYANA were just a little > to "tight" (and thus conflicting with other terms in the potential function). > So, I went back and loosened them up a bit (with CYANA's calibration routine) > and I got many fewer NOE violations, but still a lot of VDW violations. Still, > I felt this was progress. > > At this point, I had finally measured RDCs in Pf1. I started by just throwing > in the N-N RDCs along with the above described "looser" NOEs. I used one of > the example refine.py scripts (attached) and I took one of the fitted > structures from CYANA as the starting model. The calculated structures are in > good agreement with the RDCs (which was very nice to see) with tensor values > of approx -16 and 0.6 (Da and Rh). But, I still get more NOE and VDW > violations than I was expecting. > > I have to keep the identity of this protein a little quiet, but I have > attached the refine.py script and the ".xplorInput_##.sa.stats" output file. > Overall, the structures look the same as those created by CYANA, but they > aren't as precise and the geometry isn't as good. There must be something I am > doing wrong and I would love for some to point it out for me. > > Many, many thanks to anyone who read this far. > > Mike > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://dcb.cit.nih.gov/pipermail/xplor-nih/attachments/20070319/1497ede2/attachment.html
