Hi,
It is not strange that a model has a lower ProsaII Z-Score than the template, but it isn't the most common situation either. There are at least two things to keep in mind when using ProsaII Z-scores. One is that the Z-Score does not necessarily correlate with model accuracy in a short range (i.e. small differences between models are not reliably detected by ProsaII), but large differences (very good vs very poor model for the same sequence) are generally detectable. The second point is that the Prosa potentials do not "know" about anything special that your protein may have (e.g. ligands, binding sites, cofactors, etc.). We have seen Prosa produce positive (bad) energy peaks for hydrophobic binding sites as a consequence of the surface term of the Prosa energy function. In your case the Z-Score of -4.28 for a protein of 81 residues is not particularly good, the Z-Scores you get for the models ( -5.5 to -6) are more in the range of what one would expect for a good 81 residue structure. One possible explanation for the Z-Score values is that your template has a feature (like the hydrophobic binding site) that Prosa does not like, producing a lower than average Z-Score, and if that feature is absent in the models (likely since the sequence similarity is so low) it would not be surpising that if they have lower Z-scores than the template.
As to the stereochemistry part. Modeller models usually look very good in Procheck, even models based on low sequence identity. This is because Modeller optimizes the stereochemistry of the model during model building. You observe bad stereochemistry in Modeller models only when the stereochemical restraints are not compatible with the homology derived restraints, this happens sometimes with large misalignments.
I hope this helps.
Best,
Roberto
-- Roberto Sanchez, Assistant Professor Structural Biology Program, Department of Physiology & Biophysics and Institute for Computational Biomedicine, Mount Sinai School of Medicine Box 1677, 1425 Madison Avenue, New York, NY 10029 phone +1 (212) 659 8648, fax +1 (212) 849 2456 http://physbio.mssm.edu/~sanchez/
----- Original Message ----- From: "Mensur Dlakic" mensur@umich.edu To: modeller_usage@salilab.org Sent: Friday, May 17, 2002 4:40 PM Subject: Better Prosa Z-scores for models than template
> Hi, > > I have observed an interesting thing with some recently produced models. > When I evaluate them using Prosa, they all have excellent Z-scores (-5.5 to > -6 for 81 residue protein). The strange part is that the template, also 81 > residues long and solved to 2 A, has a Z-score of -4.28. When compared > using Procheck, the template and models are just about indistinguishable in > terms of stereochemistry, even though they share only 17% identity (40% > similarity). Is it "normal" that all models have better Prosa Z-scores than > the template, and what could be the reason for this "improvement"? > > Thanks for your time, > > Mensur > > =========================================================================== > | Mensur Dlakic, PhD | Tel: (734) 764-3554 | > | Department of Biological Chemistry | Fax: (734) 615-3397 | > | University of Michigan Medical School | http://www.umich.edu/~mensur/ | > | 5416 Medical Science I | http://myprofile.cos.com/mensur | > | Ann Arbor, MI 48109-0606 | E-mail: mensur@umich.edu | > =========================================================================== > > > >