Hi Denis,
I have inserted my comments in your text:
Modeller Caretaker wrote:
>Forwarding to list. > >----- Forwarded message from Denis Volkov themelon@yandex.ru ----- > >Date: Thu, 11 Sep 2003 14:15:18 +0400 >From: Denis Volkov themelon@yandex.ru >X-Mailer: The Bat! (v1.52f) >Organization: IBCH >To: modeller-care@salilab.org >Subject: Re: model_refinement > >Hello,Oliver >Thank you for fast answering! :) >No i didn't test a quality of templates course i >used 4 structures >as a templates so its difficult to do it. > You can run procheck on all 4 templates to get a feeling for the best quality you can expect from your model. Why not? How did you check the alignment. My experience is that, if the sequence alignment suggests an alignment of a segment of the target sequence with all templates, it is important that the template structures are really very similar for this segment. I.e. you need to check the alignment on basis of the superposed template structures.
>Alignment >I manually adjust >so it's not a origin of fault. I look at the model >again and think >it's not so bad as I said. But a problem is more >global: >A similarity with templates is no more than 20%. A >great deal of >mismatches are in loop origin so it's not so >bad...but can i do a good >model with so low similarity? As I said I use >Charmm, I used it to investigate >protein-protein interactions. I used Modeller model >to investigate >enzyme-substrate and enzyme-inhibitor interaction. >But after MD a >structure was a great deal disturbed. Is it a >problem of low >similarity or as you said its not good to use MD >with models maid by >Modeller? > > I would not expect too much of a model based on sequence identity with the templates in the range of 20 percent. Did you do the MD without any restraints? You could apply restraints on the regions of your model which you expect to be correct based on the alignment - in case this makes any sense with respect to the problem you are investigating. I would not expect a stable MD without restraints for a model on basis of 20% identity with the templates. But this actually is only my guess. In general one can usually not expect a homology model to get closer to the actual (experimental) structure by doing a MD simulation (see Proteins 48 (4): 593-604 SEP 1 2002).
Best, Oliver
____________________________________________________________
Oliver Hucke (PostDoc) Biomolecular Structure Center Dept. of Biochemistry University of Washington Health Sciences Building K-418C 1959 NE Pacific Ave Seattle, WA 98195 T:(206)685-7046 _______________________________