Comprehensive evaluation of protein structure alignment methods: scoring by geometric measures

We report the largest and most comprehensive comparison of protein structural alignment methods. Specifically, we evaluate six publicly available structure alignment programs: SSAP, STRUCTAL, DALI, LSQMAN, CE and SSM by aligning all 8,581,970 protein structure pairs in a test set of 2930 protein domains specially selected from CATH v.2.4 to ensure sequence diversity. We consider an alignment good if it matches many residues, and the two substructures are geometrically similar. Even with this definition, evaluating structural alignment methods is not straightforward. At first, we compared the rates of true and false positives using receiver operating characteristic (ROC) curves with the CATH classification taken as a gold standard. This proved unsatisfactory in that the quality of the alignments is not taken into account: sometimes a method that finds less good alignments scores better than a method that finds better alignments. We correct this intrinsic limitation by using four different geometric match measures (SI, MI, SAS, and GSAS) to evaluate the quality of each structural alignment. With this improved analysis we show that there is a wide variation in the performance of different methods; the main reason for this is that it can be difficult to find a good structural alignment between two proteins even when such an alignment exists. We find that STRUCTAL and SSM perform best, followed by LSQMAN and CE. Our focus on the intrinsic quality of each alignment allows us to propose a new method, called "Best-of-All" that combines the best results of all methods. Many commonly used methods miss 10-50% of the good Best-of-All alignments. By putting existing structural alignments into proper perspective, our study allows better comparison of protein structures. By highlighting limitations of existing methods, it will spur the further development of better structural alignment methods. This will have significant biological implications now that structural comparison has come to play a central role in the analysis of experimental work on protein structure, protein function and protein evolution.     

A method for simultaneous alignment of multiple protein structures

Here, we present MultiProt, a fully automated highly efficient technique to detect multiple structural alignments of protein structures. MultiProt finds the common geometrical cores between input molecules. To date, most methods for multiple alignment start from the pairwise alignment solutions. This may lead to a small overall alignment. In contrast, our method derives multiple alignments from simultaneous superpositions of input molecules. Further, our method does not require that all input molecules participate in the alignment. Actually, it efficiently detects high scoring partial multiple alignments for all possible number of molecules in the input. To demonstrate the power of MultiProt, we provide a number of case studies. First, we demonstrate known multiple alignments of protein structures to illustrate the performance of MultiProt. Next, we present various biological applications. These include: (1) a partial alignment of hinge-bent domains; (2) identification of functional groups of G-proteins; (3) analysis of binding sites; and (4) protein-protein interface alignment. Some applications preserve the sequence order of the residues in the alignment, whereas others are order-independent. It is their residue sequence order-independence that allows application of MultiProt to derive multiple alignments of binding sites and of protein-protein interfaces, making MultiProt an extremely useful structural tool.     

Comparison of conformational characteristics in structurally similar protein pairs.

Although it is known that three-dimensional structure is well conserved during the evolutionary development of proteins, there have been few studies that consider other parameters apart from divergence of the main-chain coordinates. In this study, we align the structures of 90 pairs of homologous proteins having sequence identities ranging from 5 to 100%. Their structures are compared as a function of sequence identity, including not only consideration of C alpha coordinates but also accessibility, Ooi numbers, secondary structure, and side-chain angles. We discuss how these properties change as the sequences become less similar. This will be of practical use in homology modeling, especially for modeling very distantly related or analogous proteins. We also consider how the average size and number of insertions and deletions vary as sequences diverge. This study presents further quantitative evidence that structure is remarkably well conserved in detail, as well as at the topological level, even when the sequences do not show similarity that is significant statistically.     

The PDB is a covering set of small protein structures

Structure comparisons of all representative proteins have been done. Employing the relative root mean square deviation (RMSD) from native enables the assessment of the statistical significance of structure alignments of different lengths in terms of a Z-score. Two conclusions emerge: first, proteins with their native fold can be distinguished by their Z-score. Second and somewhat surprising, all small proteins up to 100 residues in length have significant structure alignments to other proteins in a different secondary structure and fold class; i.e. 24.0% of them have 60% coverage by a template protein with a RMSD below 3.5 Å and 6.0% have 70% coverage. If the restriction that we align proteins only having different secondary structure types is removed, then in a representative benchmark set of proteins of 200 residues or smaller, 93% can be aligned to a single template structure (with average sequence identity of 9.8%), with a RMSD less than 4 Å, and 79% average coverage. In this sense, the current Protein Data Bank (PDB) is almost a covering set of small protein structures. The length of the aligned region (relative to the whole protein length) does not differ among the top hit proteins, indicating that protein structure space is highly dense. For larger proteins, non-related proteins can cover a significant portion of the structure. Moreover, these top hit proteins are aligned to different parts of the target protein, so that almost the entire molecule can be covered when combined. The number of proteins required to cover a target protein is very small, e.g. the top ten hit proteins can give 90% coverage below a RMSD of 3.5 Å for proteins up to 320 residues long. These results give a new view of the nature of protein structure space, and its implications for protein structure prediction are discussed.     

Multiple protein structure alignment.

A method was developed to compare protein structures and to combine them into a multiple structure consensus. Previous methods of multiple structure comparison have only concatenated pairwise alignments or produced a consensus structure by averaging coordinate sets. The current method is a fusion of the fast structure comparison program SSAP and the multiple sequence alignment program MULTAL. As in MULTAL, structures are progressively combined, producing intermediate consensus structures that are compared directly to each other and all remaining single structures. This leads to a hierarchic "condensation," continually evaluated in the light of the emerging conserved core regions. Following the SSAP approach, all interatomic vectors were retained with well-conserved regions distinguished by coherent vector bundles (the structural equivalent of a conserved sequence position). Each bundle of vectors is summarized by a resultant, whereas vector coherence is captured in an error term, which is the only distinction between conserved and variable positions. Resultant vectors are used directly in the comparison, which is weighted by their error values, giving greater importance to the matching of conserved positions. The resultant vectors and their errors can also be used directly in molecular modeling. Applications of the method were assessed by the quality of the resulting sequence alignments, phylogenetic tree construction, and databank scanning with the consensus. Visual assessment of the structural superpositions and consensus structure for various well-characterized families confirmed that the consensus had identified a reasonable core.     

A predicted consensus structure for the N-Terminal fragment of the heat shock protein HSP90 family

A secondary structure has been predicted for the heat shock protein HSP90 family from an aligned set of homologous protein sequences by using a transparent method in both manual and automated implementation that extracts conformational information from patterns of variation and conservation within the family. No statistically significant sequence similarity relates this family to any protein with known crystal structure. However, the secondary structure prediction, together with the assignment of active site positions and possible biochemical properties, suggest that the fold is similar to that seen in N-terminal domain of DNA gyrase B (the ATPase fragment). Proteins 27:450–458, 1997. © 1997 Wiley-Liss, Inc.     

The fragment transformation method to detect the protein structural motifs

To identify functional structural motifs from protein structures of unknown function becomes increasingly important in recent years due to the progress of the structural genomics initiatives. Although certain structural patterns such as the Asp-His-Ser catalytic triad are easy to detect because of their conserved residues and stringently constrained geometry, it is usually more challenging to detect a general structural motifs like, for example, the betabetaalpha-metal binding motif, which has a much more variable conformation and sequence. At present, the identification of these motifs usually relies on manual procedures based on different structure and sequence analysis tools. In this study, we develop a structural alignment algorithm combining both structural and sequence information to identify the local structure motifs. We applied our method to the following examples: the betabetaalpha-metal binding motif and the treble clef motif. The betabetaalpha-metal binding motif plays an important role in nonspecific DNA interactions and cleavage in host defense and apoptosis. The treble clef motif is a zinc-binding motif adaptable to diverse functions such as the binding of nucleic acid and hydrolysis of phosphodiester bonds. Our results are encouraging, indicating that we can effectively identify these structural motifs in an automatic fashion. Our method may provide a useful means for automatic functional annotation through detecting structural motifs associated with particular functions.     

Internal organization of large protein families: relationship between the sequence, structure, and function-based clustering

The protein universe can be organized in families that group proteins sharing common ancestry. Such families display variable levels of structural and functional divergence, from homogenous families, where all members have the same function and very similar structure, to very divergent families, where large variations in function and structure are observed. For practical purposes of structure and function prediction, it would be beneficial to identify sub-groups of proteins with highly similar structures (iso-structural) and/or functions (iso-functional) within divergent protein families. We compared three algorithms in their ability to cluster large protein families and discuss whether any of these methods could reliably identify such iso-structural or iso-functional groups. We show that clustering using profile-sequence and profile-profile comparison methods closely reproduces clusters based on similarities between 3D structures or clusters of proteins with similar biological functions. In contrast, the still commonly used sequence-based methods with fixed thresholds result in vast overestimates of structural and functional diversity in protein families. As a result, these methods also overestimate the number of protein structures that have to be determined to fully characterize structural space of such families. The fact that one can build reliable models based on apparently distantly related templates is crucial for extracting maximal amount of information from new sequencing projects.     

Multiple protein sequence alignment from tertiary structure comparison: assignment of global and residue confidence levels.

An algorithm is presented for the accurate and rapid generation of multiple protein sequence alignments from tertiary structure comparisons. A preliminary multiple sequence alignment is performed using sequence information, which then determines an initial superposition of the structures. A structure comparison algorithm is applied to all pairs of proteins in the superimposed set and a similarity tree calculated. Multiple sequence alignments are then generated by following the tree from the branches to the root. At each branchpoint of the tree, a structure-based sequence alignment and coordinate transformations are output, with the multiple alignment of all structures output at the root. The algorithm encoded in STAMP (STructural Alignment of Multiple Proteins) is shown to give alignments in good agreement with published structural accounts within the dehydrogenase fold domains, globins, and serine proteinases. In order to reduce the need for visual verification, two similarity indices are introduced to determine the quality of each generated structural alignment. Sc quantifies the global structural similarity between pairs or groups of proteins, whereas Pij' provides a normalized measure of the confidence in the alignment of each residue. STAMP alignments have the quality of each alignment characterized by Sc and Pij' values and thus provide a reproducible resource for studies of residue conservation within structural motifs.     

Fast structure alignment for protein databank searching.

A fast method is described for searching and analyzing the protein structure databank. It uses secondary structure followed by residue matching to compare protein structures and is developed from a previous structural alignment method based on dynamic programming. Linear representations of secondary structures are derived and their features compared to identify equivalent elements in two proteins. The secondary structure alignment then constrains the residue alignment, which compares only residues within aligned secondary structures and with similar buried areas and torsional angles. The initial secondary structure alignment improves accuracy and provides a means of filtering out unrelated proteins before the slower residue alignment stage. It is possible to search or sort the protein structure databank very quickly using just secondary structure comparisons. A search through 720 structures with a probe protein of 10 secondary structures required 1.7 CPU hours on a Sun 4/280. Alternatively, combined secondary structure and residue alignments, with a cutoff on the secondary structure score to remove pairs of unrelated proteins from further analysis, took 10.1 CPU hours. The method was applied in searches on different classes of proteins and to cluster a subset of the databank into structurally related groups. Relationships were consistent with known families of protein structure.     

MAMMOTH (matching molecular models obtained from theory): an automated method for model comparison

Advances in structural genomics and protein structure prediction require the design of automatic, fast, objective, and well benchmarked methods capable of comparing and assessing the similarity of low-resolution three-dimensional structures, via experimental or theoretical approaches. Here, a new method for sequence-independent structural alignment is presented that allows comparison of an experimental protein structure with an arbitrary low-resolution protein tertiary model. The heuristic algorithm is given and then used to show that it can describe random structural alignments of proteins with different folds with good accuracy by an extreme value distribution. From this observation, a structural similarity score between two proteins or two different conformations of the same protein is derived from the likelihood of obtaining a given structural alignment by chance. The performance of the derived score is then compared with well established, consensus manual-based scores and data sets. We found that the new approach correlates better than other tools with the gold standard provided by a human evaluator. Timings indicate that the algorithm is fast enough for routine use with large databases of protein models. Overall, our results indicate that the new program (MAMMOTH) will be a good tool for protein structure comparisons in structural genomics applications. MAMMOTH is available from our web site at http://physbio.mssm.edu/~ortizg/.     

蛋白质折叠识别方法综述

蛋白质折叠识别算法是蛋白质三维结构预测的重要方法之一,该方法在生物科学的许多方面得到卓有成效的应用。在过去的十年中,我们见证了一系列基于不同计算方式的蛋白质折叠识别方法。在这些计算方法中,机器学习和序列谱-序列谱比对是两种在蛋白质折叠中应用较为广泛和有效的方法。除了计算方法的进展外,不断增大的蛋白质结构数据库也是蛋白质折叠识别的预测精度不断提高的一个重要因素。在这篇文章中,我们将简要地回顾蛋白质折叠中的先进算法。另外,我们也将讨论一些可能可以应用于改进蛋白质折叠算法的策略。     

Conformational subspace in simulation of early-stage protein folding

A probability calculus was used to simulate the early stages of protein folding in ab initio structure prediction. The probabilities of particular phi and psi angles for each of 20 amino acids as they occur in crystal forms of proteins were used to calculate the amount of information necessary for the occurrence of given phi and psi angles to be predicted. It was found that the amount of information needed to predict phi and psi angles with 5 degrees precision is much higher than the amount of information actually carried by individual amino acids in the polypeptide chain. To handle this problem, a limited conformational space for the preliminary search for optimal polypeptide structure is proposed based on a simplified geometrical model of the polypeptide chain and on the probability calculus. These two models, geometric and probabilistic, based on different sources, yield a common conclusion concerning how a limited conformational space can represent an early stage of polypeptide chain-folding simulation. The ribonuclease molecule was used to test the limited conformational space as a tool for modeling early-stage folding.     

Assessing secondary structure assignment of protein structures by using pairwise sequence-alignment benchmarks

How to make an objective assignment of secondary structures based on a protein structure is an unsolved problem. Defining the boundaries between helix, sheet, and coil structures is arbitrary, and commonly accepted standard assignments do not exist. Here, we propose a criterion that assesses secondary structure assignment based on the similarity of the secondary structures assigned to pairwise sequence-alignment benchmarks, where these benchmarks are determined by prior structural alignments of the protein pairs. This criterion is used to rank six secondary structure assignment methods: STRIDE, DSSP, SECSTR, KAKSI, P-SEA, and SEGNO with three established sequence-alignment benchmarks (PREFAB, SABmark, and SALIGN). STRIDE and KAKSI achieve comparable success rates in assigning the same secondary structure elements to structurally aligned residues in the three benchmarks. Their success rates are between 1-4% higher than those of the other four methods. The consensus of STRIDE, KAKSI, SECSTR, and P-SEA, called SKSP, improves assignments over the best single method in each benchmark by an additional 1%. These results support the usefulness of the sequence-alignment benchmarks as a means to evaluate secondary structure assignment. The SKSP server and the benchmarks can be accessed at http://sparks.informatics.iupui.edu     

CLEMAPS: multiple alignment of protein structures based on conformational letters

CLEMAPS is a tool for multiple alignment of protein structures. It distinguishes itself from other existing algorithms for multiple structure alignment by the use of conformational letters, which are discretized states of 3D segmental structural states. A letter corresponds to a cluster of combinations of three angles formed by C(alpha) pseudobonds of four contiguous residues. A substitution matrix called CLESUM is available to measure the similarity between any two such letters. The input 3D structures are first converted to sequences of conformational letters. Each string of a fixed length is then taken as the center seed to search other sequences for neighbors of the seed, which are strings similar to the seed. A seed and its neighbors form a center-star, which corresponds to a fragment set of local structural similarity shared by many proteins. The detection of center-stars using CLESUM is extremely efficient. Local similarity is a necessary, but insufficient, condition for structural alignment. Once center-stars are found, the spatial consistency between any two stars are examined to find consistent star duads using atomic coordinates. Consistent duads are later joined to create a core for multiple alignment, which is further polished to produce the final alignment. The utility of CLEMAPS is tested on various protein structure ensembles.     

Flexible protein alignment and hinge detection

Here we present a novel technique for the alignment of flexible proteins. The method does not require an a priori knowledge of the flexible hinge regions. The FlexProt algorithm simultaneously detects the hinge regions and aligns the rigid subparts of the molecules. Our technique is not sensitive to insertions and deletions. Numerous methods have been developed to solve rigid structural comparisons. Unlike FlexProt, all previously developed methods designed to solve the protein flexible alignment require an a priori knowledge of the hinge regions. The FlexProt method is based on 3-D pattern-matching algorithms combined with graph theoretic techniques. The algorithm is highly efficient. For example, it performs a structural comparison of a pair of proteins with 300 amino acids in about 7 s on a 400-MHz desktop PC. We provide experimental results obtained with this algorithm. First, we flexibly align pairs of proteins taken from the database of motions. These are extended by taking additional proteins from the same SCOP family. Next, we present some of the results obtained from exhaustive all-against-all flexible structural comparisons of 1329 SCOP family representatives. Our results include relatively high-scoring flexible structural alignments between the C-terminal merozoite surface protein vs. tissue factor; class II aminoacyl-tRNA synthase, histocompatibility antigen vs. neonatal FC receptor; tyrosine-protein kinase C-SRC vs. haematopoetic cell kinase (HCK); tyrosine-protein kinase C-SRC vs. titine protein (autoinhibited serine kinase domain); and tissue factor vs. hormone-binding protein. These are illustrated and discussed, showing the capabilities of this structural alignment algorithm, which allows un-predefined hinge-based motions.     

Combining evolutionary information and neural networks to predict protein secondary structure

Using evolutionary information contained in multiple sequence alignments as input to neural networks, secondary structure can be predicted at significantly increased accuracy. Here, we extend our previous three-level system of neural networks by using additional input information derived from multiple alignments. Using a position-specific conservation weight as part of the input increases performance. Using the number of insertions and deletions reduces the tendency for overprediction and increases overall accuracy. Addition of the global amino acid content yields a further improvement, mainly in predicting structural class. The final network system has a sustained overall accuracy of 71.6% in a multiple cross-validation test on 126 unique protein chains. A test on a new set of 124 recently solved protein structures that have no significant sequence similarity to the learning set confirms the high level of accuracy. The average cross-validated accuracy for all 250 sequence-unique chains is above 72%. Using various data sets, the method is compared to alternative prediction methods, some of which also use multiple alignments: the performance advantage of the network system is at least 6 percentage points in three-state accuracy. In addition, the network estimates secondary structure content from multiple sequence alignments about as well as circular dichroism spectroscopy on a single protein and classifies 75% of the 250 proteins correctly into one of four protein structural classes. Of particular practical importance is the definition of a position-specific reliability index. For 40% of all residues the method has a sustained three-state accuracy of 88%, as high as the overall average for homology modelling. A further strength of the method is greatly increased accuracy in predicting the placement of secondary structure segments. © 1994 Wiley-Liss, Inc.     

Toward the solution of the protein structure prediction problem

Since Anfinsen demonstrated that the information encoded in a protein’s amino acid sequence determines its structure in 1973, solving the protein structure prediction problem has been the Holy Grail of structural biology. The goal of protein structure prediction approaches is to utilize computational modeling to determine the spatial location of every atom in a protein molecule starting from only its amino acid sequence. Depending on whether homologous structures can be found in the Protein Data Bank (PDB), structure prediction methods have been historically categorized as template-based modeling (TBM) or template-free modeling (FM) approaches. Until recently, TBM has been the most reliable approach to predicting protein structures, and in the absence of reliable templates, the modeling accuracy sharply declines. Nevertheless, the results of the most recent community-wide assessment of protein structure prediction experiment (CASP14) have demonstrated that the protein structure prediction problem can be largely solved through the use of end-to-end deep machine learning techniques, where correct folds could be built for nearly all single-domain proteins without using the PDB templates. Critically, the model quality exhibited little correlation with the quality of available template structures, as well as the number of sequence homologs detected for a given target protein. Thus, the implementation of deep-learning techniques has essentially broken through the 50-year-old modeling border between TBM and FM approaches and has made the success of high-resolution structure prediction significantly less dependent on template availability in the PDB library.