Dear Warren,
In general I agree with the idea of minimal
intervention and maximal upwards compatibility,
but that approach is producing more and more
dialect variants on the PDB format where different
people (including the BNL PDB and now the RCSB
PDB) try one bandaid after another. We now have
a format in which software can get very confused
about:
atom serial numbers -- are they unique -- one
distinct identifier for each atom -- or are they
reused as the PDB has now done for several years
in large NMR entries
atom names -- do they follow the 1992 naming
conventions that to some extent carried the
element type, or do they try to come closer
to IUPAC naming and rely on other columns
for the element name
charges -- are they there, as they are for
a few entries, or are they ignored
bonds -- are the old salt-bridge and hydrogren
bond conventions followed or are they not
followed
text -- is it all upper case or are the type-setting
conventions used
segment ids -- are they used or aren't they
etc., etc.
I believe that the compounded effect of good intentions
and scientifically sound tinkering with the PDB
format has increasingly produced bad results and
scientific errors.
It is time to stop, take a deep breath and specify
and design a workable replacement for the current
legacy format that, while it repects the needs
of existing software, avoids the trap in which
we now find ourselves of having data sets from
multiple sources that cannot be unambiguously
parsed without advice from the delphic oracle.
Regards,
Herbert J. Bernstein
=====================================================
Herbert J. Bernstein, Professor of Computer Science
Dowling College, Kramer Science Center, KSC 121
Idle Hour Blvd, Oakdale, NY, 11769
+1-631-244-3035
[EMAIL PROTECTED]
=====================================================
On Fri, 10 Aug 2007, Warren DeLano wrote:
>
> > Actually, everything proposed [will] break some software.
>
> Breaking some is far better than all, and there is a very important
> principle being followed here: The notion behind these minimalist
> improvements is that they will only break in cases where the PDB format
> itself breaks.
>
> For example, right-justifying two-letter chain IDs over columns 21-22
> means that only files that employ two-character chain IDs will break --
> otherwise you get a normal-looking one-letter chain ID in column 22.
> Likewise, the hybrid36 convention produces an otherwise normal PDB
> except when the atom count limit is exceeded. And, with the CONECT
> valence convention, if there is no valence information defined for
> HETATM ligands, then the CONECT output will be single bonds as is the
> case with standard PDB files. It is only when you have ligand valences
> defined that the output vill vary from standard PDB conventions.
>
> In other words, the file format variances only occur when they are
> needed, and that is why the proposed approaches are preferable over
> something like wide-PDB.
>
> Cheers,
> Warren
>
> > -----Original Message-----
> > From: CCP4 bulletin board [mailto:[EMAIL PROTECTED] On
> > Behalf Of Herbert J. Bernstein
> > Sent: Friday, August 10, 2007 8:53 AM
> > To: CCP4BB@JISCMAIL.AC.UK
> > Subject: Re: [ccp4bb] PDB format survey?
> >
> > Actually, everything proposed with break some software.
> > The real question is one of how much value the community
> > gains from how much of a change. mmCIF was one proposal that
> > would "solve" the problem, but which met a lot of resistance.
> > The change in atom serial numbers to strings is another
> > possibility. If you want something in between that stretches
> > the line, but preserves the programming style, take a look at:
> >
> > http://biomol.dowling.edu/WPDB/
> >
> > that extends the line and handles 999,999,999 atoms and 10
> > character chain names.
> >
> > I apologiza for the server that provides sample runs from
> > the page being down. We had a couple of bad power failures,
> > and that machine is not back in service yet, but the spec is
> > available.
> >
> > Regards,
> > Herbet J. Bernstein
> > =====================================================
> > Herbert J. Bernstein, Professor of Computer Science
> > Dowling College, Kramer Science Center, KSC 121
> > Idle Hour Blvd, Oakdale, NY, 11769
> >
> > +1-631-244-3035
> > [EMAIL PROTECTED]
> > =====================================================
> >
> > On Fri, 10 Aug 2007, Warren DeLano wrote:
> >
> > > That's easy: Backward compatibility, both in terms of old programs
> > > and old data.
> > >
> > > The idea is to maintain as much interoperability as possible.
> > >
> > > > -----Original Message-----
> > > > From: CCP4 bulletin board [mailto:[EMAIL PROTECTED]
> > On Behalf
> > > > Of Santarsiero, Bernard D.
> > > > Sent: Friday, August 10, 2007 8:17 AM
> > > > To: CCP4BB@JISCMAIL.AC.UK
> > > > Subject: [ccp4bb] PDB format survey?
> > > >
> > > > Can I ask a dumb question? Just curious...
> > > >
> > > > Why are we now limited to 80 "columns"? In the old days,
> > that was a
> > > > limit with Fortran and punched cards. Can a "record"
> > > > (whatever it's called now) be as long as we wish? Instead of
> > > > compressing a lot on a PDB record line, can we lengthen it to 130
> > > > columns?
> > > >
> > > >
> > > > Bernie Santarsiero
> > > >
> > > >
> > > > On Fri, August 10, 2007 10:10 am, Warren DeLano wrote:
> > > > > Correction: Scratch what I wrote -- the PDB format does
> > > > now support a
> > > > > formal charge field in columns 79-80 (1+,2+,1- etc.). Hooray!
> > > > >
> > > > > Thus, adoption of the CONECT valency convention is all it
> > > > would take
> > > > > for us to be able to convey chemically-defined structures using
> > > > > the PDB format.
> > > > >
> > > > > I'll happily add two-letter chain IDS and hybrid36 to PyMOL
> > > > but would
> > > > > really, really like to see valences included as well --
> > widespread
> > > > > adoption of that simple convention would represent a major
> > > > practical
> > > > > advance for interoperability in structure-based drug discovery.
> > > > >
> > > > > Cheers,
> > > > > Warren
> > > > >
> > > > >
> > > > >> -----Original Message-----
> > > > >> From: CCP4 bulletin board [mailto:[EMAIL PROTECTED]
> > > > On Behalf Of
> > > > >> Warren DeLano
> > > > >> Sent: Thursday, August 09, 2007 5:53 PM
> > > > >> To: CCP4BB@JISCMAIL.AC.UK
> > > > >> Subject: Re: [ccp4bb] PDB format survey?
> > > > >>
> > > > >> Joe,
> > > > >>
> > > > >> I feel that atom serial numbers are particularly
> > important, since
> > > > >> they, combined with CONECT records, provide the only
> > > > >> semi-standard convention I know of for reliably encoding bond
> > > > >> valences
> > > > information
> > > > >> into a PDB file.
> > > > >>
> > > > >> single bond = bond listed once
> > > > >> double bond = bond listed twice
> > > > >> triple bond = bond listed thrice
> > > > >> aromatic bond = bond listed four times.
> > > > >>
> > > > >> This is a convention long supported by tools like
> > MacroModel and
> > > > >> PyMOL.
> > > > >> For example, here is formaldehyde, where the bond between
> > > > atoms 1 and
> > > > >> 3 is listed twice:
> > > > >>
> > > > >> HETATM 1 C01 C=O 1 0.000 -0.020 0.000
> > 0.00 0.00
> > > > >> C
> > > > >> HETATM 2 N01 C=O 1 1.268 -0.765 0.000
> > 0.00 0.00
> > > > >> N
> > > > >> HETATM 3 O02 C=O 1 0.000 1.188 0.000
> > 0.00 0.00
> > > > >> O
> > > > >> HETATM 4 H01 C=O 1 1.260 -1.775 0.000
> > 0.00 0.00
> > > > >> H
> > > > >> HETATM 5 H02 C=O 1 2.146 -0.266 0.000
> > 0.00 0.00
> > > > >> H
> > > > >> HETATM 6 H03 C=O 1 -0.946 -0.562 0.000
> > 0.00 0.00
> > > > >> H
> > > > >> CONECT 1 2
> > > > >> CONECT 1 3
> > > > >> CONECT 1 3
> > > > >> CONECT 1 6
> > > > >> CONECT 2 1 4 5
> > > > >> CONECT 3 1
> > > > >> CONECT 3 1
> > > > >> CONECT 4 2
> > > > >> CONECT 5 2
> > > > >> CONECT 6 1
> > > > >>
> > > > >> I second the proposal of treating this field as a
> > unique string
> > > > >> rather than a numeric quantity.
> > > > >>
> > > > >> Two letter chain IDs would be fine with me, but I do think
> > > > we could
> > > > >> also make better use of SEGI and/or MODEL to break things up
> > > > >> while still preserving the utility of certain other records
> > > > (SHEET, HELIX,
> > > > >> etc.) within their existing column definitions.
> > > > >>
> > > > >> However, we are still lacking a standard way of designating
> > > > >> formal charges, So maybe that free column could be better used
> > > > for encoding
> > > > >> a formal charge, such as ["q" "t", "d", "-", "+", "D",
> > > > "T", "Q"] over
> > > > >> the formal charge range of [-4,-3,-2,-1,0,1,2,3,4] -- just an
> > > > >> idea :)...
> > > > >>
> > > > >> With valences plus formal charges along with expansion of
> > > > the cap on
> > > > >> atom counts, I think we could support chemically-complete
> > > > PDB files
> > > > >> and push back the date of PDB demise for a few more years!
> > > > >>
> > > > >> A Wiki dedicated to practical PDB file hacks and
> > extensions is a
> > > > >> superb idea -- of course, the goal should be to
> > ultimately come
> > > > >> up with a single well-defined standard set of hacks we all
> > > > agree upon by
> > > > >> supporting them in our code.
> > > > >>
> > > > >> Cheers,
> > > > >> Warren
> > > > >>
> > > > >> -----Original Message-----
> > > > >> From: CCP4 bulletin board [mailto:[EMAIL PROTECTED]
> > > > On Behalf Of
> > > > >> Joe Krahn
> > > > >> Sent: Thursday, August 09, 2007 1:15 PM
> > > > >> To: CCP4BB@JISCMAIL.AC.UK
> > > > >> Subject: Re: [ccp4bb] PDB format survey?
> > > > >>
> > > > >> Edward A. Berry wrote:
> > > > >> > Ethan A Merritt wrote:
> > > > >> >> On Wednesday 08 August 2007 20:47, Ralf W.
> > > > Grosse-Kunstleve wrote:
> > > > >> >>> Implementations to generate intuitive, maximally backward
> > > > >> compatible
> > > > >> >>> numbers can be found here:
> > > > >> >>>
> > > > >> >>> http://cci.lbl.gov/hybrid_36/
> > > > >> >>
> > > > >> >> From that URL:
> > > > >> >>
> > > > >> >> ATOM 99998 SD MET L9999 48.231 -64.383 -9.257 1.00
> > > > >> >> 11.54 S
> > > > >> >> ATOM 99999 CE MET L9999 49.398 -63.242 -10.211 1.00
> > > > >> >> 14.60 C
> > > > >> >> ATOM A0000 N VAL LA000 52.228 -67.689 -12.196 1.00
> > > > >> >> 8.76 N
> > > > >> >> ATOM A0001 CA VAL LA000 53.657 -67.774 -12.458 1.00
> > > > >> >> 3.40 C
> > > > >> >>
> > > > >> >> Could you please clarify this example?
> > > > >> >> Is that "A0000" a hexidecimal number, or is it a decimal
> > > > >> number that
> > > > >> >> just happens to have an "A" in front of it?
> > > > >> >> [A-Z][0-9999] gives a larger range of values than 5 bytes of
> > > > >> hexadecimal,
> > > > >> >> so I'm guessing it's the former. But the example
> > is not clear.
> > > > >> >>
> > > > >> > I'm guessing the former also. A 5-digit hex number
> > would not be
> > > > >> > backwards compatible. With this system legacy programs can
> > > > >> still read
> > > > >> > the files with 99999 atoms or less, and anything more than
> > > > >> that they
> > > > >> > couldn't have handled anyway. Very nice!
> > > > >> >
> > > > >> > Ed
> > > > >> I still prefer the idea of just truncating serial numbers,
> > > > and using
> > > > >> an alternative to CONECT for large structures.
> > > > >> Almost nobody uses atomSerial, but it still may be
> > parsed as an
> > > > >> integer, where the above idea could cause errors.
> > > > >> Furthermore, non-digit encoding still results in
> > another maximum,
> > > > >> whereas truncating the numbers has no limit. The
> > truncated serial
> > > > >> number is ambiguous only if taken out of context of the
> > > > >>
> > > > >> complete PDB file, but PDB files are by design sequential.
> > > > >>
> > > > >> Another alternative is to define an "atom-serial offset"
> > > > >> record. It can define a number which is added to all
> > subsequently
> > > > >> parsed atom serial numbers. Every ATOM/HETATM record is then
> > > > >> perfectly valid to an older program, but may only be able
> > > > to handle
> > > > >> one chunk of atoms at once.
> > > > >>
> > > > >> Likewise, I like the idea of a ChainID map record, which maps
> > > > >> single-letter chainID's to larger named ID's. Each
> > existing PDB
> > > > >> record can then be used unchanged, but files can then support
> > > > >> very long ChainID
> > > > >>
> > > > >> strings. The only disadvantage is that old PDB readers
> > will get
> > > > >> confused, but at least the individual record formats are
> > > > not changed
> > > > >> in a way that makes them crash.
> > > > >>
> > > > >> I think that keeping the old record definitions completely
> > > > unchanged
> > > > >> are
> > > > >>
> > > > >> an important feature to any PDB format revisions. Even if
> > > > we continue
> > > > >> to
> > > > >>
> > > > >> use it for another 20 years, it's primary advantage is
> > > > that it is a
> > > > >> well-established "legacy" format. If we change existing
> > > > records, we
> > > > >> break that one useful feature.
> > > > >> Therefore, I think that any changes to existing
> > records should be
> > > > >> limited to using characters positions that are currently. (The
> > > > >> one exception is that we need to make the HEADER Y2K
> > > > >>
> > > > >> compatible by using a 4-digit year, which means the existing
> > > > >> decade+year
> > > > >>
> > > > >> characters have to be moved.)
> > > > >>
> > > > >> Of course, the more important issue is that the final
> > > > decision needs
> > > > >> community involvement, and not just a decision by a
> > small group
> > > > >> of RCSB or wwPDB administrators.
> > > > >>
> > > > >> Maybe it would be useful to set up a PDB format "Wiki" where
> > > > >> alternatives can be defined, along with advantages and
> > > > disadvantages.
> > > > >> If
> > > > >>
> > > > >> there was sufficient agreement, it could be used as a
> > > > community tool
> > > > >> to put together a draft revision of the next PDB
> > format. With any
> > > > >> luck, some RCSB or wwPDB people would participate as well.
> > > > >>
> > > > >> Joe Krahn
> > > > >>
> > > > >>
> > > > >>
> > > > >>
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> >
> >
> >
> >
>
