Summary to answers to question about MR-based models generated by ROSETTA (thanks to all who replied so quickly):
***De novo methods*** 1. Arcimboldo (suggested by Isabel Uson Finkenzeller and Peter Grey) Nature Methods 6:651-3. "Crystallographic ab initio protein structure solution below atomic resolution" is based on PHASER, SHELXE and CONDOR. ***In fact this program did the job!!!*** http://chango.ibmb.csic.es/ARCIMBOLDO/ http://en.wikipedia.org/wiki/Giuseppe_Arcimboldo 2. ROSETTA and MD model comparison (by jonathan elegheert) - the smaller the protein the better. There really is an upper limit; every amino acid makes computation exponentially more intensive. - For some proteins it works better than for others. In literature, it has been suggested that the observed failures are not due to bad conformational sampling of the potential energy surface of proteins, but are rather due to the low-resolution scoring function, which sometimes fails to score the models resembling the native fold as the lowest energy structures. However, this observation is largely outweighted by the number of succesfull protein structure predictions, which make Rosetta one of the best structure prediction algorithms available to date. - In most publications, 10000 models are generated and scored. I believe something like 100000 is rather necessary. I don't know if you have access to multinode machines, but a cluster is definitely necessary. Installing the software to work with MPI is also no trivial task. - The tricky part is the scoring of the clusters of models you will generate; if you don't have the x-ray structure to score and benchmark against (since you want to use it for MR in the first place), you have to score against the most represenative structures of the largest low-energy clusters. And exactly this can be a problem (see 2nd point). - Say you'd get a model with 1.8 A rmsd with the 'real' structure (would be a very good result); this still would translate to a sequence identity of +- 30%, so already marginal. 3. ROSETTA and MRBumb (by Martyn Winn) We also had a look at this problem: D.J Rigden, R.M Keegan and M.D Winn Acta Cryst. D64 1288-1291 (2008) - "Molecular Replacement using ab initio polyalanine models generated with ROSETTA" Our approach is to use Rosetta to generate a large set of models, and feed these to MrBUMP to process. 4. Recent literature about ROSETTA (by Francois Berenger ) "High resolution protein structure prediction and the crystallographic phase problem" http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2504711/ "Prospects for de novo phasing with de novo protein models" http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2631639/ ***Different servers and model preparation tools*** 1. Alternative servers for model preparation (Colin Levy): FFAS in conjunction with either ElNEMO or CASPR 2. MR models generated by ROSETTA and other web server tools(Phyre and I-TASSER)in conjunction with MR program REM and OEDM-DEDM phase refinement tool, both included into IL MILIONE (Acta Cryst (2009) D65 477-484) (Rocco Caliandro). www.ba.ic.cnr.it 3. PYRE-SERVER (Eike Schulz) http://www.sbg.bio.ic.ac.uk/~phyre/. 4. Balbes (Edward Snell) http://www.ysbl.york.ac.uk/~fei/balbes/index.html ---------------------------------------------- Kornelius Zeth Max Planck Institute for Developmental Biology Dept. Protein Evolution Spemannstr. 35 72076 Tuebingen, Germany [email protected] Tel -49 7071 601 323 Fax -49 7071 601 349
