> > Since I know this PDB is probably a good model and it won't come up in > seq_search I was wondering how I could manually update CHAINS_all.seq > or create my own sequence database.

The latest release of MODELLER (version 7v7, released last month) has a new command called SEQFILTER that can be used to cluster PDB sequences. You can use MAKE_CHAINS (also in the latest release) to collect the PDB chains prior to running SEQFILTER.

--- Eswar Narayanan, Ph.D Mission Bay Genentech Hall 600 16th Street, Suite N474Q University of California - San Francisco San Francisco, CA 94143-2240 (CA 94158 for courier) Tel +1 (415) 514-4233; Fax +1 (415) 514-4231 http://www.salilab.org/~eashwar

It is possible to build new sequence databases for modeller - and, as Eswar said, there are two relevant commands. Writing a script to do this is unavoidable, though, unless the caretaker has one ready for everyone to download!

As a very quick fix, you could get the current pdb sequence list from here : ftp://ftp.rcsb.org/pub/pdb/derived_data/pdb_seqres.txt

Then, follow the script in modeller7v7/examples/commands/build_profile.top, which shows how you can read in a simple FASTA sequence flatfile database, like the one from the pdb website, and then use it to align against your sequence in order to build a sequence profile (and by that, retrieve all homologous sequences from the PDB).

To do the job properly, you need to apply the make_chains command (modeller7v7/examples/commands/make_chains.top) to generate the extra information that is written into the PIR information fields, and used by modeller fetch the correct PDB file for each sequence in the database.

If you have a mirror of the PDB, then this script (for unix) might work:

#!/bin/bash # makes chain records and places them pdb_seq.chn in the current working # directory. # you need to change this to point to your local copy of the PDB,

PDBDIR="/projects/biodata/pdb/data/structures/all/pdb"

for p in `ls -1 $PDBDIR` do y=`basename $p .ent.Z`; if [[ $p != $y ]]; then echo READ_MODEL FILE = '$PDBDIR/$p' > make_chains_.top echo MAKE_CHAINS MINIMAL_CHAIN_LENGTH = 30, \ MINIMAL_RESOLUTION = 2.0, MINIMAL_STDRES = 30, \ CHOP_NONSTD_TERMINII = on, \ STRUCTURE_TYPES ='structureN structureX' >> make_chains_.top mod7v7 make_chains_.top cat ${y/pdb/./}.*.chn >> pdb_seq.chn rm ${y/pdb/./}.*.chn fi done;

After that, which will take some time to run, pdb_seq.chn will contain a subset of all the PDB chains, in a similar form to the CHAINS_all.seq file.

You should, then, be able to read this new database in, apply SEQFILTER (see the example/command/seqfilter.top) , and write out the list of chain representatives (at 95%, for instance). For best use, you should rewrite the database (via READ_SEQUENCE_DB and WRITE_SEQUENCE_DB) in binary format and limit it to just the representative sequences generated by SEQFILTER (by specifying the CHAINS_LIST option on READ_SEQUENCE_DB).

Enjoy! j.

_______________________________________________________________________ Dr JB Procter:Biomolecular Modelling at ZBH - Center for Bioinformatics Hamburg http://www.zbh.uni-hamburg.de/staff.php

On Oct 3, 2004, at 7:47 AM, Bruno Afonso wrote:

> Eswar Narayanan wrote: >>> >>> Since I know this PDB is probably a good model and it won't come up >>> in seq_search I was wondering how I could manually update >>> CHAINS_all.seq or create my own sequence database. >> The latest release of MODELLER (version 7v7, released last month) has >> a new command called SEQFILTER that can be used to cluster PDB >> sequences. You can use MAKE_CHAINS (also in the latest release) to >> collect the PDB chains prior to running SEQFILTER. > > There was a mistake in my previous e-mail. :) The PDB sequence is > missing, which is *bad*, not good. I'm sorry to ask this questions, > but I'm still puzzled as to how to deal with this:

If you know exactly what your template(s) is(are) going to be, you do not have to use SEQUENCE_SEARCH to "identify" your template. You can use any of the alignment commands (ALIGN, ALIGN2D etc) to create your alignment and model your sequence based on that alignment.

> > 1) What's the criteria for make chains_all.seq? I ask this because > clearly not all of PDB is there :) and there are sequences there with > resolutions as high as 5.0 angstroms...

One usually wants to use a non-redundant version of PDB to search for templates. One way is to first select sequences of all X-ray structures that are solved at a resolution better than 3.5A, that are longer than 30aa, have no more than 10 non-standard residues, have at least 30 standard residues. These can all be specified as options to MAKE_CHAINS. You can then cluster these sequences using SEQFILTER to remove redundancies with a sequence identity threshold (usually set at 30% or 95%).

Ben has put these files on the web at http://salilab.org/modeller/supplemental.html. These are the representative sequences derived PDB files at 30% and 95% sequence identity. All x-ray and NMR PDB chains, with no limits on resolution, that are at least 30aa long, have more than 30 standard residues and not more than 10 non-standard residues were use to get these files. This is just the output of SEQFILTER on last weeks' release (09-28-04) of PDB.

> > 2) Can't I make a chains_all.seq alike with MY criteria without making > my own script? ie, is there a "right way"(TM) to do it?

See the comments above.

> > 3) I can use MAKE_CHAINS and then load the .chn as a database, but > that involves having me first finding the good PDBs that aren't on the > modeller's DB, which is kind of misses the whole point. I was using > modeller to try to find the good ones in the first place. > > Thanks for the tip on seqfilter, but my problem was the sequence > missing in the modeller's default database in the first place ;-)

The reviews listed on the modeller web-site (http://salilab.org/modeller/documentation.html) will help you understand the process of identifying a useful template for modelling.

--- Eswar Narayanan, Ph.D Mission Bay Genentech Hall 600 16th Street, Suite N474Q University of California - San Francisco San Francisco, CA 94143-2240 (CA 94158 for courier) Tel +1 (415) 514-4233; Fax +1 (415) 514-4231 http://www.salilab.org/~eashwar

On Mon, Oct 04, 2004 at 03:44:25PM +0200, J B Procter wrote: > Hi. I was wondering if there is a hard wired maximum size for the number > of sequences that can be held in memory. I got this error : > > Input/Output Error 192: Record too large

No, but there is a maximum size for a 'record', which translates to the line length. There appears to be no way round this in Fortran, short of ditching formatted IO entirely. Modeller uses a maximum line length of 16384 characters, which I guess used to be 'long enough for everyone', but some of the NR database IDs end up being longer than this, if many corrections have been made. The simplest workaround for now is to chop any excessively long lines before feeding the database to Modeller (e.g. with a Perl script). We will look into getting round this problem 'properly' in the next release.

Ben Webb, Modeller Caretaker