pymol-users

[PyMOL] alignment algorithms

[Michael Z. <mic...@ia...>]

Dear PyMOL users,

I have browsed the wiki and the first 300 mailing list messages in the
archive (the sourceforge archive does not seem to have a search set up
for the archive... arg!), but did not find any details about the
algorithms behind 'fit' and 'super.'  I understand how align works,
and it sounds like fit is align but without the processing steps
(assumes 1:1 correspondence).  Super, on the other hand, sounds more
like CE since it is sequence independent, but I cannot find any
corroboration on this.

I want to run CE or a like algorithm (sequence independent - only
concerned with spacial arrangement of alpha carbons) on a large
dataset and want to know my options in terms of speed and accuracy.

Sincerely,
Mike Z

Re: [PyMOL] alignment algorithms

[Jason V. <jas...@sc...>]

Hi Mike,

PyMOL has quite a few structure alignment and structure fitting
options.  If you want CE, check out CEalign on the PyMOLWiki
(http://www.pymolwiki.org/index.php/Cealign).  It's a script for
PyMOL.  I might build it into newer releases of PyMOL, if anyone's
interested (let me know).

"align" performs a sequence alignment and then a structure
superposition.  After that, it goes through refinement cycles to
improve the result.  Align is fast and returns quality alignments for
two sequences of high seq id.

"super" does pairwise residue alignments followed by a superposition.
It also does refinements to improve the overall quality.

"optAlign," "fit" and others are just least square fitting routines
once the atom correspondence is determined.

Regards,

-- Jason


On Wed, Jan 27, 2010 at 5:55 PM, Michael Zimmermann
<mic...@ia...> wrote:
> Dear PyMOL users,
>
> I have browsed the wiki and the first 300 mailing list messages in the
> archive (the sourceforge archive does not seem to have a search set up
> for the archive... arg!), but did not find any details about the
> algorithms behind 'fit' and 'super.'  I understand how align works,
> and it sounds like fit is align but without the processing steps
> (assumes 1:1 correspondence).  Super, on the other hand, sounds more
> like CE since it is sequence independent, but I cannot find any
> corroboration on this.
>
> I want to run CE or a like algorithm (sequence independent - only
> concerned with spacial arrangement of alpha carbons) on a large
> dataset and want to know my options in terms of speed and accuracy.
>
> Sincerely,
> Mike Z
>
> ------------------------------------------------------------------------------
> The Planet: dedicated and managed hosting, cloud storage, colocation
> Stay online with enterprise data centers and the best network in the business
> Choose flexible plans and management services without long-term contracts
> Personal 24x7 support from experience hosting pros just a phone call away.
> http://p.sf.net/sfu/theplanet-com
> _______________________________________________
> PyMOL-users mailing list (PyM...@li...)
> Info Page: https://lists.sourceforge.net/lists/listinfo/pymol-users
> Archives: http://www.mail-archive.com/pym...@li...
>



-- 
Jason Vertrees, PhD
PyMOL Product Manager
Schrodinger, LLC

(e) Jas...@sc...
(o) +1 (603) 374-7120

Re: [PyMOL] alignment algorithms

[Jason V. <jas...@sc...>]

> Dear Jason,
>
> ""super" does pairwise residue alignments followed by a superposition.
> It also does refinements to improve the overall quality."
>
> Could you elaborate a little more on this?  Do you mean that it
> calculates all pairwise distances and does a least squares fit with
> refinement steps that toss out any atoms that are more than 2 standard
> deviations from the mean residue-residue rmsd, or something more
> complex?
>
> Thank you,


Michael,

"Super" does pairwise residue comparisons using a difference of
distance matrices (as is common in other methods).  It then builds the
highest scoring alignment by taking the best scoring match, with or
without gaps, and window averaging, using a dynamic programming style
search for all possible pairs.  Think of doing a Needleman-Wunsch, but
on structural scores instead of a sequence scoring matrix (BLOSUM62).

FYI, there is an infrequently discussed "window" parameter which will
consider not just diff(atomA, atomB) but considers local substructure
as large as the window you provide:

fetch 1rlw 1rsy, async=0
super 1rlw, 1rsy, window=0

# now compare to
super 1rlw, 1rsy, window=8

This window feature is used in other algorithms for considering
average substructure, not just pairwise scores.

Try "help super" for more information on the parameters.

Hope this helps,

-- Jason




-- 
Jason Vertrees, PhD
PyMOL Product Manager
Schrodinger, LLC

(e) Jas...@sc...
(o) +1 (603) 374-7120