[Molvis-list] exporting _good_ h-bonds (was: export Chime h-bonds)

Herbert J. Bernstein yaya at bernstein-plus-sons.com
Tue Jan 3 08:53:35 EST 2006

The logic used by RasMol for H-bonds is as follows.  I would be happy
to add alternate approaches, but this one will remain the default
so as not to mess up existing scripts:

Quoting from the Sayle-Martz RasMol FAQ

"For protein, regions of alpha helix or beta-sheet are identified by
application of the algorithm of Kabsch and Sander. Hydrogen bonds are
assigned when a distance-dependent electrostatic energy of
interaction between donors and acceptors falls below a threshold
value. No reference is made to phi or psi angles.

"For DNA, Watson-Crick double helix internucleotide bonding rules are
applied without reference to actual distances between donors and
acceptors. This can lead to gross errors. For example, in 1d66.pdb,
cytosine 28 was positioned erroneously (Ronen Marmorstein, personal
communication). Two impossible hbonds are drawn, disregarding the
actual geometry. One hbond is drawn from G11.N2 to C28.O2 across one
intervening bond and one intervening atom, a distance of 7 Angstroms!
(Hbond donor-acceptor distances are generally less than 3.5

The detailed code is available in molecule.c


Look under the heading:

/*  Kabsch & Sander DSSP Structure Determination  */

The choice of drawing H-bonds between the donor and acceptor atoms of
the hydrogen bond or between C-alpha atoms of a protein backbone or
between phosphrous atoms of a nucleic acid backbone is made
by use of the

   set hbonds sidechain
   set hbonds backbone


The choice of drawing H-bonds only between atoms of the same chain
or of allowing interchain H-bonds is made by the command

   set hbonds chain false

for intra-chain H-bonds only, and

   set hbonds chain true

for inter-chain H-bonds

At 4:32 PM -0800 1/2/06, Raymond Keller wrote:
>* Kevin Karplus (karplus at soe.ucsc.edu) [20060102 14:32]:
>>  I'm not familiar with the programs available, because I wrote my own
>>  for the undertaker program.  It turns out to be fairly tricky to get
>>  "real" hbonds, and there is always an arbitrary cutoff for whatever
>>  properties are measured (distances, angles, ...).  There are also
>>  questions about whether weak Hbonds (such as donor hydrogens on
>>  araomatic carbons, or acceptors in the center of an aromatic ring)
>>  should be counted.
>>  I would be a bit uncomfortable using someone else's code, unless I
>>  spent many hours reading it carefully and seeing how all the arbitrary
>>  decisions had been made.
>I should clarify my situation.
>I'm a commercial contractor responsible for rendering client
>representations of proteins into different forms, including physical
>media.  My needs are not academic, and as such they are not subject
>to the ostensibly ardent dedication to truth that should typify
>academic research.  Interestingly, while the academic search for
>truth may be tainted by the need for securing grants, my primarily
>commercial efforts are tainted by a personal desire to be accurate.
>Clients get to dictate secondary structure assignments as suits
>them.  I find it odd that many don't seem concerned about how the
>determinations are made.  I've come to assume that the "standards"
>are in flux (if glacially so).
>A current client is interested in having his RasMol representation
>processed, including h-bonds, and this explains my request for all
>y'all's advice.
>Here the taint of desire for accuracy comes into play:  If I can't
>get RasMol's data, I may opt for an alternative.  I have been able
>to hack Stride's sources to make it output its internal h-bond
>information (normally it only outputs residues).  Presumably a
>program that uses phi and psi values should be more accurate than
>one that uses bond distance alone.  (I'm having a hard time finding
>a description of the method(s) employed by RasMol, so the
>insinuation of method should be taken as slander until proven.)  I
>am but a simple trained primate.  Your advanced biomolecular science
>frightens and confuses me.  Still, the Stride sources appear to take
>advantage of torsion angles (and empirically-derived hydrogen bond
>energy)--necessarily so, it seems, for the computation of "Baker"
>and "Rose" h-bonds.  From this and my assumptions about RasMol I
>believe that Stride's output is of higher quality, and will use this
>alternative for my client, barring his refusal.
>Do you think it's a better choice?
>Raymond Scott Keller
>independent molecular visualization contractor
>Santa Cruz, California, United States
>Molvis-list mailing list
>Molvis-list at bioinformatics.org

  Herbert J. Bernstein, Professor of Computer Science
    Dowling College, Kramer Science Center, KSC 121
         Idle Hour Blvd, Oakdale, NY, 11769

               Office:  +1-631-244-3035
            Lab (KSC 020): +1-631-244-3451
                  yaya at dowling.edu

More information about the Molvis-list mailing list