Tableau-based protein substructure search using quadratic programming

Software

Source code, scripts: qptabsearch.tar.gz

You can also get the source code and databases from the old site http://munk.cis.unimelb.edu.au/~stivalaa/qpprotein/ (which, however, I can no longer update), and the source code from GitHub.

This software can be used freely for any purpose, modified, redistributed, etc. with no restrictions. However we would appreciate it if you acknowledge your use of it, and in particular if you would cite our paper in any publication that makes use of it.

QP Tableau Search is also available for online use via the Pro-origami web server.

Tableau databases

The tableau databases were built with buildtableauxdb.py -t dssp -3 -5 -p none and converted to the ASCII format available here with convdbnumeric2ascii.py and convdbpacked2ascii.py for the numeric and tableaux formats respectively. The distance matrices were built with buildtableauxdb.py -t dssp -3 -5 -p none -d and the combined tableaux and distance matrix database in ASCII format created with the convdb2.py script. These scripts are included with the source code.

Combined tableau and distance matrix database for ASTRAL SCOP 1.73 95% sequence identity subset

Tableau database for ASTRAL SCOP 1.73 95% sequence identity subset

Numeric database for ASTRAL SCOP 1.73 95% sequence identify subset

Randomly permuted tableaux

In the "Non-linear matchings" section of our paper, we describe the use of random permutations of tableaux as an artificial test to verify the capability of our method to find non-linear matchings, i.e. sets of correspondences between SSEs in which the sequential order of corresponding SSEs is not preserved. Here we provide some more details on this method for those who might like to use to evaluate other non-linear structural alignment techniques.

As described in the paper (and citations therein) a tableau is a square symmetric matrix in which each row represents the orientation of one SSE relative to every other SSE. Normally, the rows (and columns) are in the same order as the SSEs in the protein sequence, from N to C terminus. In order to generate randomly permuted tableaux, we instead generate the tableaux by randomly permuting the sequence of SSEs, so that rather than the rows being ordered according to the sequence, they are ordered according to the random permutation that was generated.

The -u option on the pytableaucreate.py script included with the source code is used to perform this task. It also outputs the permutation used, which is necessary so that any alignment based on the permuted tableau can be mapped back to the actual SSEs in the structure which the row and column in the tableau represents. The ssepermutationremap.py script accomplishes this task, and is used by the qptabmatchstructs.sh script when the -u option is used to randomly permute the query structure. This script illustrates the pipeline used to perform an alignment (possibly with a permuted tableau) and visualize it using PyMOL. The genpermutedqueries.sh script was used to generate the queries described in the paper.

All of the scripts mentioned here are provided with the source code download, and all contain internal documentation as to their purpose and usage.

Faster implementation

As described in our paper, all the results published there were with tsrchd_sparse the sparse matrix implementation of the QP solver using the UMFPACK solver in SuiteSparse. However, if you have the Intel Math Kernel Library (MKL) version 10.1, which contains the PARDISO sparse linear solver, you can use the tsrchd_pardiso implementation, which is approximately 70% faster than tsrchd_sparse.

Reference

If you use our software, data, or results in your research, please cite:

Stivala, A., Wirth, A. and Stuckey, P., Tableau-based protein substructure search using quadratic programming BMC Bioinformatics 2009, 10:153

Contact

Alex Stivala