Ben,
Great response! Indeed, the PDB's limited capacity for secondary
structure does indeed represent this as one single helix, not two.
> +/- 15 degrees is pretty difficult to see - so I think that DSSPs
> +assignment
> in this case, while liberal, is certainly acceptable.
I hardly think it acceptable that one residue with 3-10 PHI/PSI
constitutes a helix when you have so many other telltale characteristics
of a Type II' beta turn, including a Glycine in position i+1 and all
deviations within 22 deg. of ideal turn, as well as four planer CA atoms
(actually five, C-alphas 98-102 are all nearly planer). Some turns do
look like helices, but not this one -- not even close.
> Great example of where DSSP fails, but perhaps a bit contrived.
I don't think this example is contrived. One of the many tests
I put PYMOL through regularly is to render the entire PDB over several
hours, so I literally see thousands of structures on a routine basis.
Such "objective but bogus" assignments are far more common than one
would like to think. It is a real problem.
> The reason that consistency is of paramount importance in assigning
secondary structure
> is so that different scientists can talk about the same protein in
context.
> If everyone uses the same definition of secondary structure - "helix
1" is always "helix 1".
Good point, but I disagree. In acknowledging the subjective and
relative nature of assignments, I think it is more reasonable to ask
scientists to define their terms rather than to perpetuate broken
conventions. Given how evolution, crystallization, and protein
processing work, even with DSSP, Helix 3 can easily be Helix 2 or 4 in
another structure of the identical or a related protein, so the savings
in confusion is slight. Also, keep in mind that all numbering is
relative and increasingly problematic as more and more homologous
structures are solved, and explicit conceptual labels are far more
general than numeric ones.
But finally, do understand that I would provide DSSP as an
option if it were possible to do so, but frankly, it's not. The paper
does not enable an exact reimplementation, and the owners of DSSP,
instead of making DSSP source code usable and redistributable as an open
standard, have decided to use the software for revenue generation
(presumed modest at best). It is exatly this kind of short-sighted
restrictiveness that PyMOL was created to combat. We must end this
dismal pattern of scientists developing research software that is so
drastically and artificially limited by licensing terms that it has
little or no ability to enable future scientific advances.
Computer-dependent scientific fields like structural biology and
computational chemistry will be handicapped until we can all effectively
access and improve upon existing software systems. That will not happen
until research scientists and academic institutions stop viewing such
software programs as potentially lucrative intellectual properties and
start viewing them as the critical building blocks that will enable
future breakthroughs if and when they become widely shared.
Imagine... What would happen if all scientists stopped
publishing their discoveries and kept all information within one lab or
institution? Scientific progress would come to a halt. That's
basically what has occurred with so much scientific software over the
last 20 years. Don't get me wrong -- some great code has been written
-- and some of that code has been shared without cost to some or for a
low fee to others. However, most of that code has almost never been
shared in such a way as to enable extension, modification, and
redistribution. As a result, we are not making much cumulative
progress. Software programs resemble ideas: they must be shared,
questioned, altered, and refined in order for lasting progress to occur.
Otherwise they stagnate just like old ways of thinking.
DSSP (circa 1983) is a great example of how many
computer-dependent sciences remain trapped in the 1980s. Whether it
works better than PyMOL or not is almost beside the point. It is simply
not an appropriate standard for us to be using today in this age of open
standards, open source, and composite systems. We can and should do
better than this.
My hope is that if PyMOL and enough similar projects are
successful, that governments, companies, and funding agencies will begin
to back these kinds of efforts wholeheartedly with money and policies,
and thus drive rapid progress and widespread benefits throughout the
sciences.
Cheers,
Warren
--
mailto:war...@delanoscientific.com
Warren L. DeLano, Ph.D.
Principal Scientist
DeLano Scientific LLC
Voice (650)-346-1154
Fax (650)-593-4020
> -----Original Message-----
> From: Ben Hitz [mailto:bh...@exelixis.com]
> Sent: Monday, November 03, 2003 11:01 AM
> To: Warren L. DeLano; pymol-users@lists.sourceforge.net
> Subject: Re: [PyMOL] DSSP
>
>
> Warren -
>
> Just up front I want to state that I don't think DSSP is the
> best possible way to algorithmically determine secondary
> structure. My position is that secondary structure is a
> subjective assignment anyway - so it will always be open to
> interpretation. In that case, you are better off assigning
> to the standard this is in most common usage.
>
> Great example of where DSSP fails, but perhaps a bit contrived.
>
> First of all, DSSP is assigning this as a "H" helix and "G"
> 3-10 helix - not really a single continuous helix (although
> one could interpret it this way). Second, you point about it
> not really being a 3-10 helix is not strong..
>
> 3-10 helix (-49/-26). (ideal)
> residue 101 (-63/-12)
>
> +/- 15 degrees is pretty difficult to see - so I think that DSSPs
> +assignment
> in this case, while liberal, is certainly acceptable.
>
> Finally the difference between a 3 residue "helix" and a
> reverse turn is not really easy to quantitate. In fact, any
> single turn of any helix is basically, well, a turn. Any
> algorithm that "allows" 3-4 residue helices is going to make
> such assignments, and in fact any realistic (human)
> interpretation is going to make the equivalence of short
> "helices" with turns.
>
> Your last point in the URL is critical. There will
> presumably be "errors" in dss, or stride, or any other
> algorithm, because frankly, the problem is not very well
> defined. Hence, it's probably best to just implement
> something, and leave the interpretation to the user. The
> reason that consistency is of paramount importance in
> assigning secondary structure is so that different scientists
> can talk about the same protein in context. If everyone uses
> the same definition of secondary structure - "helix 1" is
> always "helix 1".
>
> Ben
>
>
> --
> Ben Hitz Exelixis
> Computational Biology and Informatics
> bh...@exelixis.com 650-837-8137
> ----- Original Message -----
> From: "Warren L. DeLano" <war...@delanoscientific.com>
> To: "'Ben Hitz'" <bh...@exelixis.com>;
> <pymol-users@lists.sourceforge.net>
> Sent: Monday, November 03, 2003 10:39 AM
> Subject: RE: [PyMOL] DSSP
>
>
> > Ben,
> >
> > Granted -- and there are certain times when it makes sense for
> > everyone to be "wrong" in order to be consistent. But is this one
> > such case? I think not. If DSSP "helices" aren't actually
> helices,
> > then of what use is the standard? Neither the resulting statistics
> > nor the cartoon figures will be correct.
> >
> > I've put up one particularly egregious DSSP example on the web at
> >
> > http://www.pymol.org/not_helix
> >
> > for inspection. If you are comfortable with obvious turn regions
> > (such as this one) being assigned as helix, then by all means, go
> > right ahead and continue using DSSP via rTools or the PDB.
> >
> > But for me, I am not going to lose sleep over deviating from an
> > existing standard in PyMOL. As I said before, the DSSP
> paper wasn't
> > even written in such a way as to enable exact reproduction of their
> > research other than by running their proprietary (1000
> EURO) software
> > program or by analyzing their source code.
> >
> > PyMOL's new "dss" algorithm is nowhere near validated
> enough to become
> > a standard, but I do think that if we are going to agree upon a
> > standard for secondary structure assignment, that the
> standard should
> > reflect common notions of secondary structure, be exactly
> described,
> > be reproducible, and be available in an open-source reference
> > implementation. Does DSSP meet these criteria? No, but
> STRIDE might
> > (if one could find it), and "dss" probably could in time.
> >
> > Cheers,
> > Warren
> >
> > --
> > mailto:war...@delanoscientific.com
> > Warren L. DeLano, Ph.D.
> > Principal Scientist
> > DeLano Scientific LLC
> > Voice (650)-346-1154
> > Fax (650)-593-4020
> >
> > > -----Original Message-----
> > > From: pymol-users-ad...@lists.sourceforge.net
> > > [mailto:pymol-users-ad...@lists.sourceforge.net] On Behalf Of Ben
> > > Hitz
> > > Sent: Monday, November 03, 2003 8:28 AM
> > > To: pymol-users@lists.sourceforge.net
> > > Subject: [PyMOL] DSSP
> > >
> > >
> > > Warren -
> > > The advantage of using DSSP over another heuristic
> method, is that
> > > it is a standard. A DSSP helix is a DSSP helix
> everywhere, even if a
> > > crystallographer might extend the end a bit.
> > >
> > > Ben
> > > --
> > > Ben Hitz Exelixis
> > > Computational Biology and Informatics
> > > bh...@exelixis.com 650-837-8137
> > >
