[Owl-devel] reconstruct

Jose M. Duarte jose.m.duarte at gmail.com
Thu Jan 26 03:58:19 EST 2012


Hi Ioannis

Well I'm impressed that you could compile tinker, nice! and also glad to
have a "customer" after all this time ;)

So your questions one by one:

1) cns (xplor-cns). I never tried it but it was the thing I always wanted
to do... I think it is much better than tinker and anyway it is the
standard for NMR these days, so if anything at least you get structures
that are similar to NMR structures in PDB. I guess it's not complicated to
write an adapter for it with the code we have... if I only had the time! ;)

2) tinker 6. Didn't try either, but the dynamic memory allocation is THE
feature that we needed. Did you try compiling it? or even using their
provided binaries with reconstruct? That in principle should solve all
problems we have with Tinker <5

3) reconstruct output: rec.pdb is the elongated pdb (use as input for
distgeom), rec.report is the report with rmsds and so on so this one you
should look at. Otherwise the important ones are the final models
rec.xxx.pdb

4) other parameters: -e you should definitely use, otherwise you get crap
models that are not even accepted by the CASP server. For -f (force
constant) I'd use the default. BTW don't use fast mode, always use
simulated annealing

5) third column: reconstruct ignores it as long as I remember. It wouldn't
be difficult to implement assigning different force constants for different
edge weights. But what we don't know is what kind of mapping
weight-to-force-constant is reasonable... that part would require quite
some time to optimise

6) sampling: our rule was 40 models minimum (anyway half of them are
mirrors), but of course the more the better. Indeed there's a random seed
for each model, so in principle there's no difference 1x20 vs 10x2. One
thing I'd recommend to use is the cluster feature (-A). That was working
quite well and one gets linear scaling (one model per cpu)

Well that's about it. Let me know how it goes, especially if you try tinker
6. I'd be interested to know if that works at all.

Cheers

Jose





On 25 January 2012 13:46, Ioannis Filippis <filippis at gmail.com> wrote:

> Hi guys,
>
> I have been mailing already Jose about reconstruction.
>
> I managed to compile tinker 5.1.09 and run reconstruction-1.0 and worked
> fine!!! The only problem with compiling in 64 (when increasing the
> parameters in sizes.i) was some error message "Relocation truncated to fit"
> that I overcame using -mcmodel=medium in compiling and linking.
>
> I have a few questions:
> -First, what all these files in output dir are? I guess I only need the
> rec.XXX.pdb files as the predicted models
> ld-mjeste16 109% ls test/
> rec.001  rec.001.pdb  rec.key  rec.pdb  rec.report  rec.seq
> rec.tinker.log  rec.xyz
> -Second, what parameters do you usually use? I guess all default settings
> but I am not sure about -e and -f
> -Third, does tinker take into account the last column in cm file? So if I
> have contact predictions with confidence/probabilities, will this be taken
> into account? If not, any simple way to enforce it?
> -How many models have to be generated to get a good sampling of output?
> Any difference between running reconstruct once and generating 20 models or
> running it 10 times and generating 2models each time (some seed somewhere)?
> -Any approximate run times?
>
> I guess I have heard all of these before but don't really remember!
>
> Many thanks!
>
> Best,
> Ioannis
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.bioinformatics.org/pipermail/owl-devel/attachments/20120126/5f4c3a9b/attachment.html>


More information about the Owl-devel mailing list