Assessing secondary structure assignments of protein structures by using pairwise sequence-alignment benchmarks, the secondary structure of a protein refers to the local conformation of its polypeptide backbone. Knowing secondary structures of proteins iessential for their structure classification, understanding folding dynamics and mechanis(s)ms, and discovering conserved structural/functional motifs. Secondary structure informatiis also useful for sequence and multiple sequence alignment, structure alignment,andsequ(on)ence to structure alignment (or threading). As a result, predicting secondary structures from protein sequences continues to be an active field of research fifty six years after Pauling and Corey first predicted that the most common regular patterns of proteibackbones are the α-helix and the β-sheet. Prediction and application of proteinsecon(n)dary structures rely on prior assignment of the secondary-structure elements from a given protein structure by human or computational methods.

   Many computational methods have been developed to automate the assignment of secondary structures. Examplare DSSP,STRIDE, DEFINE, P-SEA, KAKSI,P-CURVE, XTLSSTR,SECSTR,SEGNO,(es)and VoTAP. These methods are based on either the hydrogen-bond pattern, geometric features, expert knowledge or their combinations. However, they often disagree on their assignments. For example, disagreement among DSSP, P-CURVE, and DEFINE can be as large as 25%. More beta sheet is assigned by XTLSSTR and more pi-helix by SECSTR than by DSSP. The discrepancy among different methods is caused by non-ideal configurations of helices and sheets. As a result, defining the boundaries between helix, sheet, and coil is problematical and a significant source of discrepancies between different methods.

   Inconsistent assignment of secondary structures by different methods highlights the need for a criterion or a benchmark of “standard” assignments that could be used to assess and compare assignment methods. One possibility is to use the secondary structures assigned by the authors who solved the protein structures. STRIDE, in fact, has been optimized to achieve the highest agreement with the authors’ annotations. However, it is not clear what is the criterion used for manual or automatic assignment of secondary structures by different authors. Another possibility is to treat the consensus prediction by several methods as the gold standard. However, there is no obvious reason why each method should weight equally in assigning secondary structures and which method should be used in consensus. Other used criteriinclude helix-capping propensity, the deviation from ideal helical and sheet configurations,and(a)structural accuracy produced by sequence-to-structure alignment guided by secondary structure assignment.

   In this paper, we propose to use sequence-alignment benchmarks for assessing secondary structure assignments. These benchmarks are produced by 3D-structure alignment of structurally homologous proteins. Instead of assessing the accuracy of secondary-structure assignment directly, which is not yet feasible, we compare the two assignments of secondary structures in structurally aligned positions. We assume that the best method should assign the same secondary-structure element to the highest fraction of structurally aligned positions. Certainly, structurally aligned positions do not always have the same secondary structures. Moreover, different structure-alignment methods do not always produce the same lt. Nevertheless, this criterion provides a means to locate a secondary-structure assig(resu)nment method that is most consistent with tertiary structure alignment. We suggest that this approach provides an objective evaluation of secondary structure assignment methods.

上一页  [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]  ... 下一页  >>