Reference

If you use the results of the CS23D2.0 server in a publication, please cite the following paper:
Wishart DS, Arndt D, Berjanskii M, Tang P, Zhou J, Lin G. CS23D: a web server for rapid protein structure generation using NMR chemical shifts and sequence data. Nucleic Acids Res. 2008 Jul 1;36(Web Server issue):W496-502.

Tables

Table 1: RMSDs of distorted X-ray structures before and after chemical shift refinement. The high quality X-ray structures of 9 proteins, for which complete chemical shifts were also available, were distorted by randomly varying all phi and psi angles uniformly within ±5°. The distorted structures were then minimized using chemical shift refinement with GAFolder for 300 iterations. Corrected chemical shifts from RefDB were used.

Table 2: Results of modeling and refining 9 structures and comparison with native X-ray and NMR structures. Native structures were excluded from CS23D2.02.0 databases during model generation. Refinement for each model involved 10 runs of 300 iterations each, and selection of the model with the lowest energy.

# Rama violations - indicates the number of residues that fall in the generously allowed and disallowed regions of the Ramachandran plot as adapted from Lovell SC, et al. Structure validation by Calpha geometry: phi,psi and Cbeta deviation. Proteins 2003 Feb 15;50(3):437-50. The fewer violations, the better the structure. Ideally there should be no violations.

# Omega violations - indicates the number of residues that have Omega angles that are not in the range 180.0 ± 12.5 degrees or 0.0 ± 15.0 degrees. Ideal Omega angles for trans peptide bonds are 180 degrees while Ideal Omega angles for cis peptide bonds are 0 degrees.

Mean chemical shift correlation - indicates the Pearson correlation coefficient calculated between input chemical shifts and those calculated by SHIFTX for the structure generated by CS23D2.0. Values closer to 0.9 are optimal. Values less than 0.7 indicate possible problems with the structure.

HCount (%) - indicates the percentage of residues in the protein that are involved in canonical hydrogen bonds. Good structures have higher percentages of "good" or canonical hydrogen bonds. Values approaching 80% are ideal. To ensure consistency, hydrogen atoms for all structures were replaced by those from Reduce v3.10 (Word, et al.(1999) "Asparagine and glutamine: using hydrogen atom contacts in the choice of sidechain amide orientation" J. Mol. Biol. 285, 1733-1745).

Average H-bond energy - indicates the average H-bond energy calculated in kcal/mol (using the DSSP definitions of Kabsch W, Sander C. Dictionary of protein secondary structure: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers. 1983 Dec;22(12):2577-637). Average H-bond energy indicates how close H-bonds are to ideal geometry. The lower the average H-bond energy, the better the structure is.

Table 3. Performance of CS23D2.0 in generating 3D structures for 62 randomly selected BMRB files. The given RMSDs are those of the final structure relative to the known (X-ray or NMR) structure.

Table 4. Backbone RMSD between CS23D2.0 models and corresponding experimental structures for recent entries in BMRB database (as of the first week of April 2008).

Table 5. Backbone RMSDs between models generated by CS23D2.0 sub-program for ab initio folding and corresponding experimental structures.

Table 6: Ubiquitin models built with CS23D2.0 using (a) homology modelling, (b) threading, and (c) ab initio folding. RMSDs and sequence identity values were calculated with respect to the native structure (1UBQ). Tables (a) and (b) show the ubiquitin models built as ubiquitin homologs were progressively removed from the database, so that the homolog used in a given row is the best one found by CS23D2.0 when the more closely matching homologs above it were removed. (c) Results of ab initio folding of ubiquitin by Rosetta. 50 models were produced in each of 10 runs. Backbone RMSDs of the model with the best CS23D2.0 score are shown, with those less than 2.8 Å labeled as successes. Homologous proteins were excluded from Rosetta's fragment generation step.

(a) Homology Modelling

(b) Threading

(c) Ab initio

Table 7. Comparison of model global accuracy achieved by CS23D2.0 and similar programs.



(a). Comparison of backbone RMSDs of models generated by CS-ROSETTA, CHESHIRE, and CS23D2.0.

(b) Comparison of backbone RMSDs of models generated by CS-ROSETTA and CS23D2.0 for proteins that did not meet CS-ROSETTA convergence criteria.

(c) Comparison of backbone RMSDs of models generated by CS-ROSETTA and CS23D2.0 for structural genomics proteins.

Figure 1. Dependence of backbone RMSD of CS23D2.0 models on template sequence identity for different methods of structure generation

Energy Function

Summary

Score Weighting

CS23D2.0 Results Assessment

   0.75 - 1.00 = High
   0.65 - 0.75 = Good
   0.55 - 0.65 = Moderate
   0.00 - 0.55 = Poor

   #res in phi/psi core          90%
   #res in phi/psi allowed        7%
   #res in phi/psi generous       1%
   #res in phi/psi disallowed     0%
   #res in omega allowed         99%
   #res in omega disallowed       1%

Torsion Angle Mapping for Thrifty