Motivation: The size of current protein databases is a challenge for many Bioinformatics applications, both in terms of processing speed and information redundancy. It may be therefore desirable to efficiently reduce the database of interest to a maximally representative subset.
Results: The MinSet method employs a combination of a Suffix Tree and a Genetic Algorithm for the generation, selection and assessment of database subsets. The approach is generally applicable to any type of string-encoded data, allowing for a drastic reduction of the database size whilst retaining most of the information contained in the original set. We demonstrate the performance of the method on a database of protein domain structures encoded as strings. We used the SCOP40 domain database by translating protein structures into character strings by means of a structural alphabet and by extracting optimized subsets according to an entropy score that is based on a constant-length fragment dictionary. Therefore, optimized subsets are maximally representative for the distribution and range of local structures. Subsets containing only 10% of the SCOP structure classes show a coverage of >90% for fragments of length 1-4.
Availability: http://mathbio.nimr.mrc.ac.uk/~jkleinj/MinSet.
Supplementary information: Supplementary data are available at Bioinformatics online.