Code for defining canonical classes

Reads a set of PDB files and a control file which specifies the loops to be analysed. For each loop writes a file containing the CA pseudo-torsions for the loops. Shorter loops are padded with torsions of 9999.0

For each loop in turn, Performs cluster analysis. Process results of C.A. to identify a sensible number of clusters. Assigns each conformation to a cluster. Look for features in each cluster?? Conserved buried residues? Cluster analysis must provide the centre of the cluster, the dimensions of the cluster and the distance to the nearest other cluster such that new structures can be scanned against the clusters.

Three programs:

1. CLAN - CLuster ANalysis.

Performs cluster analysis on a loop in a set of PDB files. Generates information on the clusters. Takes an input file of the following syntax:

METHOD    <clustering method>           ! Ward, single, multiple, etc.
LOOP      <pdb> <startres> <lastres>    ! Multiple records
OUTPUT    <outfile>                     ! or stdout if not specified
MAXLENGTH <length>                      ! Max length of a loop in analysis
SCHEME    <insert scheme>               ! Order in which positions in the 
                                        ! loop should be assigned from the 
                                        ! actual loop i.e. where insertions 
                                        ! should be considered
DENDOGRAM                               ! Show the clustering tree dendogram
TABLE                                   ! Show the cluster table
DATA                                    ! Show the data which is used for
                                        ! clustering
POSTCLUSTER <cutoff>                    ! Specify RMS cutoff for post-cluster

The output file contains the METHOD, MAXLENGTH and SCHEME information as well as the clustering data which includes the centre and size of each cluster and the distance to the nearest neighbouring cluster.

2. FICL - FInd CLuster

Takes a loop in a PDB file and matches it against a set of clusters to find which cluster (if any) it matches.

Is run with the following syntax:

ficl <datafile> <pdb> <startres> <lastres>

where

<pdb> <startres> <lastres>    is the loop to be tested
<datafile>                    is the output file from CLAN

FICL will pick up the cluster method from the output of CLAN

CLAN must be run with TABLE and DATA switched on!

3. FINDSDRS - Find Structurally Determining Residues

Analyzes the output of CLAN to identify the key residues responsible for defining the canonical class.

Run with:

findsdrs [-k] [clanfile [outfile]]
         -k Keep intermediate solvent accessibility files so they
            don't have to be recalculated for the next run.

Note that solvent accessibility calculation requires pdbsolv from BiopTools to be in the path.

1. Automation

• Ensure that you can carry cluster labels forward from one run of the code to the next. There are probably scripts to do this already somewhere here, but they are not documented and I don't know where they are!

• Generate a file in the correct format for our canonical identification program

2. Check the clusters

• Are the clusters really sensible? Could we improve them by using different clustering thresholds? For example, 12E8 CDR-L1 and 1A0Q CDR-L1 are both in the same cluster, but have a backbone flip at positions L30-L31. Should and could these be split into two clusters?

3. CDR-H3

• CDR-H3 doesn't really form canonicals. However in our old paper we found that there were some clusters for 7-residue CDR-H3s. Extend this work and see if this holds up and whether we can now do longer CDR-H3s.

• Most of this is automated by the doit.sh script in the demo directory

1. Get a list of antibody structures (SACS: /acrm/www/html/abs/sacs/antibodies.xml or preferably from Jake's database) (You need to write this)
2. All the PDB files live in /acrm/data/pdb/pdbXXXX.ent Apply standard Chothia numbering to each PDB file extracting just the numbered Fv fragment and storing them in a directory (We should have a script that does this already)
3. For each of the CDRs, create an input file for CLAN (see below) and run the CLAN program - generates the sets of clusters and tells you which PDB files fall into each cluster. (You need to automate the creation of the input files for CLAN)
4. For each CDR, run the findsdrs program. This generates a list of the key residues which define each of the clusters. (This should simply be a case of running the program on the CLAN output)

• These stages are not yet automated

5. You need to maintain some sort of historical mapping so that you have cluster names (e.g. '11A') that are maintained in a consistent way. This needs to look at the CLAN output and check that antibodies that were assigned to a particular class stay in a class with that name. (You need to code this).
6. You need to convert the output from findsdrs into an input file for our canonicals program that assigns canonical class on the basis of a sequence file. (You need to code this).
7. Run the complete pipeline on the ~700 available antibody structures to update the results we had on ~50 antibodies.

Seen by Tanner