BCSearch: fast structural fragment mining over large collections of protein structures

Resources to mine the large amount of protein struc-tures available today are necessary to better un-derstand how amino acid variations are compatible with conformation preservation, to assist protein de-sign, engineering and, further, the development of biologic therapeutic compounds. BCSearch is a ver-satile service to efficiently mine large collections of protein structures. It relies on a new approach based on a Binet–Cauchy kernel that is more discrimina-tive than the widely used root mean square devia-tion criterion. It has statistics independent of size even for short fragments, and is fast. The system-atic mining of large collections of structures such as the complete SCOPe protein structural classifica-tion or comprehensive subsets of the Protein Data Bank can be performed in few minutes. Based on this new score, we propose four innovative applica-tions: BCFragSearch and BCMirrorSearch, respec-tively, search for fragments similar and anti-similar to a query and return information on the diversity of the sequences of the hits. BCLoopSearch identifies candidate fragments of fixed size matching the flanks of a gaped structure. BCSpecificitySearch analyzes a complete protein structure and returns information about sites having few similar fragments. BCSearch is available a