1
|
So you've analyzed your NOE data sets and have 1000+ assigned restraints, with probably inevitable errors. Going through each one of them is extremely tedious and even by looking at the assignment it's not always easy to tell whether it's right or wrong. How do you find those errors most efficiently? What clues do you use to detect errors in the NOE assignment? |
|
3
|
I am assuming that you are dealing with NOE data for biomacromolecules such as proteins, as it is unlikely that you will have such a large number of assigned restraints in other systems. Simplest method is possibly a bootstrap method (used commonly in statistical analysis): leave out a small fraction e.g. 2-5 % of the NOe derived restraints and determine the 3D structure. If the resulting structure predicts short distances consistent with your original data then the excluded NOe based restraints are validated. The process can be repeated with different subsets of excluded NOe restraints. A second possibly strategy is to use relaxation matrix analysis to identify likely contributions from spin diffusion and eliminate/correct the corresponding restraints. A third method is to use distance geometry. If the metric matrix has only 3 non zero eigenvalues then all the NOe derived restraints are selfconsistent. If a few of them are incorrect, and if these incorrect restraints imply structural inconsistency in 3D then more then 3 eigenvalues will have magnitudes that are significantly different from nonzero ( you will find more details regarding this in the publications of havel and wuthrich; havel and crippen; etc ). Dear Sekhar, yes "leave some out" validation works, perhaps the only disadvantage of this method is that there are very many ways to select a sub-group of restraints from a long list. So the algorithm wont scale well. - Evgeny Fadeev (Mar 02 '10 at 08:42) |
|
2
|
What software package are you using for your calculations? Some programs should automatically generate a list of violations in your NOE and / or assignment lists. I use NIH-XPLOR. The thing is that violated restraint is not necessarily the one that creates the problem, so it's not always productive to just look at the violated restraints. Thanks for joining the forum, btw!!! I will lower the barrier for commenting and voting soon to bootstrap the community. - Evgeny Fadeev (Feb 22 '10 at 09:46) |
|
0
|
I've tried QUEEN approach in the past (when I was a postdoc in Rob Clubb's lab at UCLA) and found that it helps to identify important restraints quickly. With the list of restraints sorted by the order of importance one can start with validating the most significant restraints first. This allows concentrating on a subset of maybe 10% of all restraints. QUEEN stands for QUantitative Evaluation of Experimental Nmr restraints. When you look at the sorted list of restraints you'll quickly notice that long range restraints are on the top of the list, quite predictable. However QUEEN method is quantitative - it's based on the analysis of information content hidden in the set of distance restraints (or other geometrical restraints that can be mapped into plain distance restraints). At the time when I tried the software it had a limitation of working only with the proteins. linksbtw, the post is styled blue because it is by the author of the question - maybe not the greatest choice of color and style ;) - Evgeny Fadeev (Mar 02 '10 at 08:50) |
