Automatic comparison and classification of protein structures.

The classification and alignment of multiple three-dimensional protein structures is a powerful way to detect similarities that cannot be discovered from the sequences alone and can help to infer phylogeny. However, the alignment process remains problematic for divergent structures. We have devised a fully automatic pipeline, HSF, drawing its inspiration from well-known structural alignment methods, which given a list of structures not only aligns all pairs but also classifies them fully. We demonstrate proof of principle for the new method by aligning the currently available set of highly diverged virus coat protein structures containing double β-barrels, as well as validating the method with established test sets for multiple structural alignments. The results for the virus proteins are inline with previous observations based on biochemical, genetic and structural studies but go further, since by providing coherent alignments between sets of molecules with marked structural distortion, they facilitate the marshaling of arguments for or against homology. The classification results can therefore be readily interpreted in terms of phylogeny.