Accuracy analysis of multiple structure alignments

Abstract

Protein structure alignment methods are essential for many different challenges in protein science, such as the determination of relations between proteins in the fold space or the analysis and prediction of their biological function. A number of different pairwise and multiple structure alignment (MStA) programs have been developed and provided to the community. Prior knowledge of the expected alignment accuracy is desirable for the user of such tools. To retrieve an estimate of the performance of current structure alignment methods, we compiled a test suite taken from literature and the SISYPHUS database consisting of proteins that are difficult to align. Subsequently, different MStA programs were evaluated regarding alignment correctness and general limitations. The analysis shows that there are large differences in the success between the methods in terms of applicability and correctness. The latter ranges from 44 to 75% correct core positions. Taking only the best method result per test case this number increases to 84%. We conclude that the methods available are applicable to difficult cases, but also that there is still room for improvements in both, practicability and alignment correctness. An approach that combines the currently available methods supported by a proper score would be useful. Until then, a user should not rely on just a single program.

MRFy: Remote Homology Detection for Beta-Structural Proteins Using Markov Random Fields and Stochastic Search

“…It accomplishes this by allowing flexibility in the form of small, geometrically impossible bends and breaks in a protein structure, in order to distort that structure into alignment with another protein. Matt was shown to perform particularly well compared to competing multiple and pairwise structure alignment programs on proteins whose homology was similar to the SCOP superfamily level [MBC08,RSWD09,BSL09]. Surprisingly, we find that our automatic classification scheme based on a pairwise distance value derived from Matt, coupled with a straightforward neighbor-joining algorithm to construct the hierarchical clusters [SMP08] matches SCOP better than previous automatic methods, at the superfamily, and even, to some extent, at the fold level.…”

Comparing the intrinsic dynamics of multiple protein structures using elastic network models

“…Comparing distant homologues provides a challenge in defining which parts of the proteins to compare. This is commonly solved by structural alignment, which is a challenging problem, particularly for the simultaneous alignment of sets of proteins [70][71][72][73].…”

DALIX: Optimal DALI Protein Structure Alignment

“…We limited the length to maximally 50 residues to obtain alignments for which our algorithm can explore multiple branch-and-bound nodes within a few minutes. SISY and RIPC [23], [25] are data sets of manually curated structural alignments assembled from the Sisyphus collection [32], which are difficult for alignment programs because of repetitions, large indels, circular permutations, conformational variability, and so on. The consolidated SISY and RIPC sets consist of 98 and 22 alignments, respectively.…”

“…The consolidated SISY and RIPC sets consist of 98 and 22 alignments, respectively. With consolidated, we denote the subsets that have been consulted for evaluation in [25].…”

“…For this purpose, we use 1) alignments of SCOPCath domains [24] with lengths between 30 and 50 residues who share family, superfamily, or fold; 2) alignments from the SKOLNICK data set, which has been first used in [13] and contains alignment instances of proteins from the same family; and 3) alignments from the SISY and RIPC collections [23], [25]. We find that DALI is very reliable in returning a good alignment according to DALI scoring.…”