Hi Margot

Without wishing to sound egoistic, you might want to take a look at some of my modelling articles. The iterative approach to modelling is exactly the one I take. Basically, I make an initial alignment of my target with one or more structures, based on fold recognition (often) adjusted using common sense so that insertions and deletions are well accommodated. I then make a SET of models - 15 or 20, employing a large (4A) randomisation in modeller. In this way I try to explore lots of conformational possibilities. I then analyse the set with PROSA II (freely available) which gives me 1) an overall score that can support or not the fold recognition [there's the pG server too] and 2) profiles which highlight problematic bits of the models. I then look at the 'bad' bits to see if I can imagine any alignment variants there that would produce better models (more typically packed and better solvent exposure). I then make new sets of models with these new ideas and see if they're better than the original set. And so on, until I can't make any more improvements.

Assuming that PROSA II can distinguish between poorer and better models (and there's cause for optimism - in the original paper it can pick up threading errors in crystal structures) then the final model should be better than the original.

I also do PROCHECK analysis, mainly towards the end. I used to use VERIFY_3D, but then we upgraded machines and I no longer have access. Very frustratingly, this program seems unavailable. There's a server for individual structures but that's no use for people working with sets.

Best wishes

Daniel

+-------------------------------------------------------------------------+ | Dr Daniel John Rigden | | CENARGEN/EMBRAPA | e-mail: daniel@cenargen.embrapa.br | | Parque Estacao Biologica | http://www.cenargen.embrapa.br | | PqEB - Final - Av. W3 Norte | Phone: +55 (61)448-4741 | | 70770-900, Brasilia-D.F.-BRAZIL | Fax: +55 (61)340-3658 | +-------------------------------------------------------------------------+

On Thu, 4 Jul 2002, Margot Ernst wrote:

> > Hi Modeller developers and friends, > > in Sanchez and Sali 1997 Curr Opin Struct Biol 7:206-214 > I read that a possible way to improve on homology modeling below 30 PID could > be "iterative changes in the alignment during caluclation" > My question is: Is this a great idea that has been realized already and I > just missed where and how, is it work in progress or has it been discarded > again due to technical problems? > Could one, alternatively, if only one template is availble, use this template > several times with different alignment variants (home-made version of > modeling on several templates, so to speak) so that different homology > derived restraints go into the same VTF, or is it, at present knowledge, > better to go for ensembles of models and then try to construct one or more > "best" consensus models? > Any and all information and ideas are much appreciated! > margot > > -- > Dr. Margot Ernst > Abteilung Prof. Sieghart > Institut f. Hirnforschung > Spitalgasse 4 > A-1090 WIEN > margot.ernst@univie.ac.at > visit brainresearch: www.univie.ac.at/brainresearch > > >