Rebuilding sequence databases

Older versions of the Modeller distribution contain a sequence database, in the files `modlib/CHAINS_*`. These files are

• `CHAINS_all.seq` or `pdball.pir`: sequences for every chain in every structure in the PDB.
• `CHAINS_3.0.95_XN.cod` or `pdb_95.cod`: all chains are clustered at 95% sequence identity, and for each cluster, the PDB code of the representative chain is listed in this file.
• `CHAINS_3.0.95_XN.grp` or `pdb_95.grp`: for each representative, the other chains which are 95% sequence identical.
• `CHAINS_3.0.40_XN.cod` and `CHAINS_3.0.40_XN.grp`: similar files, clustered at 40% sequence identity.

These files are obviously not updated whenever the PDB is, and are not included at all with newer versions of Modeller, but you can download updated copies from our supplemental data file download page, or regenerate them yourself if you have a local copy of PDB. Firstly, to build `CHAINS_all.seq` or `pdball.pir`:

1. For each PDB file, run a script similar to that below. Set `code` to the PDB code and set `atom_files_directory` to the directory containing your local copy of PDB:

#!python
from modeller import *

e = Environ()
e.io.atom_files_directory = ['/database/pdb/']

code = '1xyz'
m = Model(e, file=code)

m.make_chains(file=code, minimal_chain_length=30, minimal_stdres=30,
              chop_nonstd_termini=True, max_nonstdres=10,
              minimal_resolution=99.0, structure_types='structureN structureX')

1. This will produce a `.chn` file for every chain in PDB. Concatenate these together (e.g. with the Unix `cat` command) to make the new `CHAINS_all.seq` or `pdball.pir` file.

Now you can build the `.cod` and `.grp` files for any sequence identity cutoff using the following script (adjust the `seqid_cut` variable accordingly):

#!python
from modeller import *

e = Environ()

s = SequenceDB(e, seq_database_file='CHAINS_all.seq', chains_list='all',
               seq_database_format='PIR', minmax_db_seq_len=(30, 3000),
               clean_sequences=True)

s.filter(matrix_offset=-450, rr_file='${LIB}/blosum62.sim.mat',
         gap_penalties_1d=(-500, -50), seqid_cut=40,
         output_grp_file='CHAINS_3.0.40_XN.grp', output_cod_file='CHAINS_3.0.40_XN.cod')

Note that this will take a long time to run. For high (>90%) sequence identity cutoffs, it is more efficient to use CD-HIT instead. A script that automates this is included as part of ModPipe (`python/ClusterPDB.py`).