Hi,
In general using more than one template for model building will improve
the quality of the final models. But this is true only if the different
templates are complementary in the sense that some regions of one
template are closer to the target structure than the corresponding
regions in another template and vice versa. This will generally happen
when you have several templates which are at a similar sequence identity
distance from the target but are also different from each other. For
example having two templates which are 40% and 45% identical to the
target sequence but are only 35% identical to each other is a very good
situation.
In your case you seem to have one high sequence identity template and
several low sequence identity templates. It is unlikely that the low
sequence identity templates will improve the model you would obtain by
using only the high sequence identity template. Of course this depends
on the particular case and it could be that a low sequence identity
contributes something positive to the model, but it is unlikely. As a
general rule it is better to build the model only on the higher sequence
identity template and other templates that have similar sequence
identity (given they are significantly different from each other as
explained above).
With respect to the alignment of the structures: you can superpose the
structures of all the templates on each other and deduce their alignment
from this superposition using the MALIGN3D command in MODELLER.
I hope this helps.
Best Wishes,
Roberto
--
Roberto Sanchez | phone : (212) 327 7206
The Rockefeller University | fax : (212) 327 7540
1230 York Avenue, Box 38 | e-mail:
New York, NY 10021-6399 | http://salilab.org
Gulshat Ibragimova wrote:
>
> Dear Modellers,
>
> what could be the best strategy for building models (especially when one
> has to build lot of them -:)? The "search_sequence" procedure gives a list
> of proteins with similar sequences. However, their structures are
> generally misaligned. So this very diverse structural information might
> lead to bad models. Perhaps, it would make sense to find out structural
> relatives of the most identical protein (using the FSSP, at example) and
> to use them for building models despite their (relatives') low sequence
> similarity?
>
> Thanking for any comments,
> Gulshat
>
> ---------------------------------------------------------------------
> Gulshat Ibragimova, Ph.D.
> EMBL, Postfach 102209
> 69012 Heidelberg, FRG
> Tel: +49 (0)6221 387466, Fax: +49 (0)6221 387517