3D-Hit: fast structural comparison of proteins on multicore architectures

Abstract

3D-Hit is a well established method for rapid detection of structural similarities between proteins, which is widely used in various bioinformatics web servers (MetaServer, GRDB, 3D-Fun, Rosetta, etc.). The algorithm decomposes proteins into set of overlaping segments of 9–13 residues, then tries to match them using root mean square distance metric. The best aligned pairs of segments are selected as seeds for futher analysis. Those initial hits are expanded by iterative process in order to construct the global structural alignment by concatenating pairs of matching segments. The method has the same accuracy as the other state-of-the-art structural comparison algorithms (LGscore2, DALI), yet it provides much faster processing times, and can be used in a high-throughput setup as the structural module of bioinformatics pipelines. The method is optimized in terms of speed and accuracy to work on novel computer architectures, such as PowerXCell8i and Sun Constellation System. Here, we provide the source code of the 3D-Hit program, describe selected architectures on which the software was ported, present programing models, point out significant porting steps and sumarize performance comparisons.