Exploring the effects of sparse restraints on protein structure prediction

One of the main barriers to accurate computational protein structure prediction is searching the vast space of protein conformations. Distance restraints or inter‐residue contacts have been used to reduce this search space, easing the discovery of the correct folded state. It has been suggested that about 1 contact for every 12 residues may be sufficient to predict structure at fold level accuracy. Here, we use coarse‐grained structure‐based models in conjunction with molecular dynamics simulations to examine this empirical prediction. We generate sparse contact maps for 15 proteins of varying sequence lengths and topologies and find that given perfect secondary‐structural information, a small fraction of the native contact map (5%‐10%) suffices to fold proteins to their correct native states. We also find that different sparse maps are not equivalent and we make several observations about the type of maps that are successful at such structure prediction. Long range contacts are found to encode more information than shorter range ones, especially for α and αβ‐proteins. However, this distinction reduces for β‐proteins. Choosing contacts that are a consensus from successful maps gives predictive sparse maps as does choosing contacts that are well spread out over the protein structure. Additionally, the folding of proteins can also be used to choose predictive sparse maps. Overall, we conclude that structure‐based models can be used to understand the efficacy of structure‐prediction restraints and could, in future, be tuned to include specific force‐field interactions, secondary structure errors and noise in the sparse maps.     

Long-range information and physicality constraints improve predicted protein contact maps

Protein topology representations such as residue contact maps are an important intermediate step towards ab initio prediction of protein structure, but the problem of predicting reliable contact maps is far from solved. One of the main pitfalls of existing contact map predictors is that they generally predict unphysical maps, i.e. maps that cannot be embedded into three-dimensional structures or, at best, violate a number of basic constraints observed in real protein structures, such as the maximum number of contacts for a residue. Here, we focus on the problem of learning to predict more "physical" contact maps. We do so by first predicting contact maps through a traditional system (XXStout), and then filtering these maps by an ensemble of artificial neural networks. The filter is provided as input not only the bare predicted map, but also a number of global or long-range features extracted from it. In a rigorous cross-validation test, we show that the filter greatly improves the predicted maps it is input. CASP7 results, on which we report here, corroborate this finding. Importantly, since the approach we present here is fully modular, it may be beneficial to any other ab initio contact map predictor.     

Ab initio and template-based prediction of multi-class distance maps by two-dimensional recursive neural networks

Background

Prediction of protein structures from their sequences is still one of the open grand challenges of computational biology. Some approaches to protein structure prediction, especially ab initio ones, rely to some extent on the prediction of residue contact maps. Residue contact map predictions have been assessed at the CASP competition for several years now. Although it has been shown that exact contact maps generally yield correct three-dimensional structures, this is true only at a relatively low resolution (3–4 Å from the native structure). Another known weakness of contact maps is that they are generally predicted ab initio, that is not exploiting information about potential homologues of known structure.

Results

We introduce a new class of distance restraints for protein structures: multi-class distance maps. We show that C _α trace reconstructions based on 4-class native maps are significantly better than those from residue contact maps. We then build two predictors of 4-class maps based on recursive neural networks: one ab initio, or relying on the sequence and on evolutionary information; one template-based, or in which homology information to known structures is provided as a further input. We show that virtually any level of sequence similarity to structural templates (down to less than 10%) yields more accurate 4-class maps than the ab initio predictor. We show that template-based predictions by recursive neural networks are consistently better than the best template and than a number of combinations of the best available templates. We also extract binary residue contact maps at an 8 Å threshold (as per CASP assessment) from the 4-class predictors and show that the template-based version is also more accurate than the best template and consistently better than the ab initio one, down to very low levels of sequence identity to structural templates. Furthermore, we test both ab-initio and template-based 8 Å predictions on the CASP7 targets using a pre-CASP7 PDB, and find that both predictors are state-of-the-art, with the template-based one far outperforming the best CASP7 systems if templates with sequence identity to the query of 10% or better are available. Although this is not the main focus of this paper we also report on reconstructions of C _α traces based on both ab initio and template-based 4-class map predictions, showing that the latter are generally more accurate even when homology is dubious.

Conclusion

Accurate predictions of multi-class maps may provide valuable constraints for improved ab initio and template-based prediction of protein structures, naturally incorporate multiple templates, and yield state-of-the-art binary maps. Predictions of protein structures and 8 Å contact maps based on the multi-class distance map predictors described in this paper are freely available to academic users at the url http://distill.ucd.ie/.     

Protein fold recognition and dynamics in the space of contact maps

We introduce an energy function for contact maps of proteins. In addition to the standard term, that takes into account pair-wise interactions between amino acids, our potential contains a new hydrophobic energy term. Parameters of the energy function were obtained from a statistical analysis of the contact maps of known structures. The quality of our energy function was tested extensively in a variety of ways. In particular, fold recognition experiments revealed that for a fixed sequence the native map is identified correctly in an overwhelming majority of the cases tested. We succeeded in identifying the structure of some proteins that are known to pose difficulties for such tests (BPTI, spectrin, and cro-protein). In addition, many known pairs of homologous structures were correctly identified, even when the two sequences had relatively low sequence homology. We also introduced a dynamic Monte Carlo procedure in the space of contact maps, taking topological and polymeric constraints into account by restrictive dynamic rules. Various aspects of protein dynamics, including high-temperature melting and refolding, were simulated. Perspectives of application of the energy function and the method for structure checking and fold prediction are discussed. Proteins 26:391–410 © 1996 Wiley-Liss, Inc.     

Protein homology detection and fold inference through multiple alignment entropy profiles

Homology detection and protein structure prediction are central themes in bioinformatics. Establishment of relationship between protein sequences or prediction of their structure by sequence comparison methods finds limitations when there is low sequence similarity. Recent works demonstrate that the use of profiles improves homology detection and protein structure prediction. Profiles can be inferred from protein multiple alignments using different approaches. The "Conservatism-of-Conservatism" is an effective profile analysis method to identify structural features between proteins having the same fold but no detectable sequence similarity. The information obtained from protein multiple alignments varies according to the amino acid classification employed to calculate the profile. In this work, we calculated entropy profiles from PSI-BLAST-derived multiple alignments and used different amino acid classifications summarizing almost 500 different attributes. These entropy profiles were converted into pseudocodes which were compared using the FASTA program with an ad-hoc matrix. We tested the performance of our method to identify relationships between proteins with similar fold using a nonredundant subset of sequences having less than 40% of identity. We then compared our results using Coverage Versus Error per query curves, to those obtained by methods like PSI-BLAST, COMPASS and HHSEARCH. Our method, named HIP (Homology Identification with Profiles) presented higher accuracy detecting relationships between proteins with the same fold. The use of different amino acid classifications reflecting a large number of amino acid attributes, improved the recognition of distantly related folds. We propose the use of pseudocodes representing profile information as a fast and powerful tool for homology detection, fold assignment and analysis of evolutionary information enclosed in protein profiles.     

Prediction of probable pathways of folding in globular proteins

A method is described for the prediction of probable folding pathways of globular proteins, based on the analysis of distance maps. It is applicable to proteins of unknown spatial structure but known amino acid sequence as well as to proteins of known structure. It is based on an objective procedure for the determination of the boundary of compact regions that contain high densities of interresidue contacts on the distance map of a globular protein. The procedure can be used both with contact maps derived from a known three-dimensional protein structure and with predicted contact maps computed by means of a statistical procedure from the amino acid sequence alone. The computed contact map can also be used to predict the location of compact short-range structures, viz. -helices and -turns, thereby complementing other statistical predictive procedures. The method provides an objective basis for the derivation of a theoretically predicted pathway of protein folding, proposed by us earlier [Tanaka and Scheraga (1977) Macromolecules10, 291–304; Némethy and Scheraga (1979) Proc. Natl. Acad. Sci., U.S.A.76, 6050–6054].     

Reconstruction of 3D structures from protein contact maps

The prediction of the protein tertiary structure from solely its residue sequence (the so called Protein Folding Problem) is one of the most challenging problems in Structural Bioinformatics. We focus on the protein residue contact map. When this map is assigned it is possible to reconstruct the 3D structure of the protein backbone. The general problem of recovering a set of 3D coordinates consistent with some given contact map is known as a unit-disk-graph realization problem and it has been recently proven to be NP-Hard. In this paper we describe a heuristic method (COMAR) that is able to reconstruct with an unprecedented rate (3-15 seconds) a 3D model that exactly matches the target contact map of a protein. Working with a non-redundant set of 1760 proteins, we find that the scoring efficiency of finding a 3D model very close to the protein native structure depends on the threshold value adopted to compute the protein residue contact map. Contact maps whose threshold values range from 10 to 18 Ångstroms allow reconstructing 3D models that are very similar to the proteins native structure.     

Toward an accurate prediction of inter-residue distances in proteins using 2D recursive neural networks

Background

Protein inter-residue contact maps provide a translation and rotation invariant topological representation of a protein. They can be used as an intermediary step in protein structure predictions. However, the prediction of contact maps represents an unbalanced problem as far fewer examples of contacts than non-contacts exist in a protein structure. In this study we explore the possibility of completely eliminating the unbalanced nature of the contact map prediction problem by predicting real-value distances between residues. Predicting full inter-residue distance maps and applying them in protein structure predictions has been relatively unexplored in the past.

Results

We initially demonstrate that the use of native-like distance maps is able to reproduce 3D structures almost identical to the targets, giving an average RMSD of 0.5Å. In addition, the corrupted physical maps with an introduced random error of ±6Å are able to reconstruct the targets within an average RMSD of 2Å. After demonstrating the reconstruction potential of distance maps, we develop two classes of predictors using two-dimensional recursive neural networks: an ab initio predictor that relies only on the protein sequence and evolutionary information, and a template-based predictor in which additional structural homology information is provided. We find that the ab initio predictor is able to reproduce distances with an RMSD of 6Å, regardless of the evolutionary content provided. Furthermore, we show that the template-based predictor exploits both sequence and structure information even in cases of dubious homology and outperforms the best template hit with a clear margin of up to 3.7Å. Lastly, we demonstrate the ability of the two predictors to reconstruct the CASP9 targets shorter than 200 residues producing the results similar to the state of the machine learning art approach implemented in the Distill server.

Conclusions

The methodology presented here, if complemented by more complex reconstruction protocols, can represent a possible path to improve machine learning algorithms for 3D protein structure prediction. Moreover, it can be used as an intermediary step in protein structure predictions either on its own or complemented by NMR restraints.     

Information and discrimination in pairwise contact potentials

We examine the information-theoretic characteristics of statistical potentials that describe pairwise long-range contacts between amino acid residues in proteins. In our work, we seek to map out an efficient information-based strategy to detect and optimally utilize the structural information latent in empirical data, to make contact potentials, and other statistically derived folding potentials, more effective tools in protein structure prediction. Foremost, we establish fundamental connections between basic information-theoretic quantities (including the ubiquitous Z-score) and contact "energies" or scores used routinely in protein structure prediction, and demonstrate that the informatic quantity that mediates fold discrimination is the total divergence. We find that pairwise contacts between residues bear a moderate amount of fold information, and if optimized, can assist in the discrimination of native conformations from large ensembles of native-like decoys. Using an extensive battery of threading tests, we demonstrate that parameters that affect the information content of contact potentials (e.g., choice of atoms to define residue location and the cut-off distance between pairs) have a significant influence in their performance in fold recognition. We conclude that potentials that have been optimized for mutual information and that have high number of score events per sequence-structure alignment are superior in identifying the correct fold. We derive the quantity "information product" that embodies these two critical factors. We demonstrate that the information product, which does not require explicit threading to compute, is as effective as the Z-score, which requires expensive decoy threading to evaluate. This new objective function may be able to speed up the multidimensional parameter search for better statistical potentials. Lastly, by demonstrating the functional equivalence of quasi-chemically approximated "energies" to fundamental informatic quantities, we make statistical potentials less dependent on theoretically tenuous biophysical formalisms and more amenable to direct bioinformatic optimization.     

Correlated mutations and residue contacts in proteins

The maintenance of protein function and structure constrains the evolution of amino acid sequences. This fact can be exploited to interpret correlated mutations observed in a sequence family as an indication of probable physical contact in three dimensions. Here we present a simple and general method to analyze correlations in mutational behavior between different positions in a multiple sequence alignment. We then use these correlations to predict contact maps for each of 11 protein families and compare the result with the contacts determined by crystallography. For the most strongly correlated residue pairs predicted to be in contact, the prediction accuracy ranges from 37 to 68% and the improvement ratio relative to a random prediction from 1.4 to 5.1. Predicted contact maps can be used as input for the calculation of protein tertiary structure, either from sequence information alone or in combination with experimental information. © 1994 John Wiley & Sons, Inc.     

Statistical potential-based amino acid similarity matrices for aligning distantly related protein sequences

Aligning distantly related protein sequences is a long-standing problem in bioinformatics, and a key for successful protein structure prediction. Its importance is increasing recently in the context of structural genomics projects because more and more experimentally solved structures are available as templates for protein structure modeling. Toward this end, recent structure prediction methods employ profile-profile alignments, and various ways of aligning two profiles have been developed. More fundamentally, a better amino acid similarity matrix can improve a profile itself; thereby resulting in more accurate profile-profile alignments. Here we have developed novel amino acid similarity matrices from knowledge-based amino acid contact potentials. Contact potentials are used because the contact propensity to the other amino acids would be one of the most conserved features of each position of a protein structure. The derived amino acid similarity matrices are tested on benchmark alignments at three different levels, namely, the family, the superfamily, and the fold level. Compared to BLOSUM45 and the other existing matrices, the contact potential-based matrices perform comparably in the family level alignments, but clearly outperform in the fold level alignments. The contact potential-based matrices perform even better when suboptimal alignments are considered. Comparing the matrices themselves with each other revealed that the contact potential-based matrices are very different from BLOSUM45 and the other matrices, indicating that they are located in a different basin in the amino acid similarity matrix space.     

Prediction of contact maps by GIOHMMs and recurrent neural networks using lateral propagation from all four cardinal corners

MOTIVATION: Accurate prediction of protein contact maps is an important step in computational structural proteomics. Because contact maps provide a translation and rotation invariant topological representation of a protein, they can be used as a fundamental intermediary step in protein structure prediction. RESULTS: We develop a new set of flexible machine learning architectures for the prediction of contact maps, as well as other information processing and pattern recognition tasks. The architectures can be viewed as recurrent neural network implemantations of a class of Bayesian networks we call generalized input-output HMMs (GIOHMMs). For the specific case of contact maps, contextual information is propagated laterally through four hidden planes, one for each cardinal corner. We show that these architectures can be trained from examples and yield contact map predictors that outperform previously reported methods. While several extensions and improvements are in progress, the current version can accurately predict 60.5% of contacts at a distance cutoff of 8 A and 45% of distant contacts at 10 A, for proteins of length up to 300.     

Analysis and prediction of functional sub-types from protein sequence alignments

The increasing number and diversity of protein sequence families requires new methods to define and predict details regarding function. Here, we present a method for analysis and prediction of functional sub-types from multiple protein sequence alignments. Given an alignment and set of proteins grouped into sub-types according to some definition of function, such as enzymatic specificity, the method identifies positions that are indicative of functional differences by comparison of sub-type specific sequence profiles, and analysis of positional entropy in the alignment. Alignment positions with significantly high positional relative entropy correlate with those known to be involved in defining sub-types for nucleotidyl cyclases, protein kinases, lactate/malate dehydrogenases and trypsin-like serine proteases. We highlight new positions for these proteins that suggest additional experiments to elucidate the basis of specificity. The method is also able to predict sub-type for unclassified sequences. We assess several variations on a prediction method, and compare them to simple sequence comparisons. For assessment, we remove close homologues to the sequence for which a prediction is to be made (by a sequence identity above a threshold). This simulates situations where a protein is known to belong to a protein family, but is not a close relative of another protein of known sub-type. Considering the four families above, and a sequence identity threshold of 30 %, our best method gives an accuracy of 96 % compared to 80 % obtained for sequence similarity and 74 % for BLAST. We describe the derivation of a set of sub-type groupings derived from an automated parsing of alignments from PFAM and the SWISSPROT database, and use this to perform a large-scale assessment. The best method gives an average accuracy of 94 % compared to 68 % for sequence similarity and 79 % for BLAST. We discuss implications for experimental design, genome annotation and the prediction of protein function and protein intra-residue distances.     

MANIFOLD: protein fold recognition based on secondary structure, sequence similarity and enzyme classification

We present a protein fold recognition method, MANIFOLD, which uses the similarity between target and template proteins in predicted secondary structure, sequence and enzyme code to predict the fold of the target protein. We developed a non-linear ranking scheme in order to combine the scores of the three different similarity measures used. For a difficult test set of proteins with very little sequence similarity, the program predicts the fold class correctly in 34% of cases. This is an over twofold increase in accuracy compared with sequence-based methods such as PSI-BLAST or GenTHREADER, which score 13-14% correct first hits for the same test set. The functional similarity term increases the prediction accuracy by up to 3% compared with using the combination of secondary structure similarity and PSI-BLAST alone. We argue that using functional and secondary structure information can increase the fold recognition beyond sequence similarity.     

1001 optimal PDB structure alignments: integer programming methods for finding the maximum contact map overlap.

Protein structure comparison is a fundamental problem for structural genomics, with applications to drug design, fold prediction, protein clustering, and evolutionary studies. Despite its importance, there are very few rigorous methods and widely accepted similarity measures known for this problem. In this paper we describe the last few years of developments on the study of an emerging measure, the contact map overlap (CMO), for protein structure comparison. A contact map is a list of pairs of residues which lie in three-dimensional proximity in the protein's native fold. Although this measure is in principle computationally hard to optimize, we show how it can in fact be computed with great accuracy for related proteins by integer linear programming techniques. These methods have the advantage of providing certificates of near-optimality by means of upper bounds to the optimal alignment value. We also illustrate effective heuristics, such as local search and genetic algorithms. We were able to obtain for the first time optimal alignments for large similar proteins (about 1,000 residues and 2,000 contacts) and used the CMO measure to cluster proteins in families. The clusters obtained were compared to SCOP classification in order to validate the measure. Extensive computational experiments showed that alignments which are off by at most 10% from the optimal value can be computed in a short time. Further experiments showed how this measure reacts to the choice of the threshold defining a contact and how to choose this threshold in a sensible way.     

Support Vector Machines for Protein Fold Class Prediction

Knowledge of the three‐dimensional structure of a protein is essential for describing and understanding its function. Today, a large number of known protein sequences faces a small number of identified structures. Thus, the need arises to predict structure from sequence without using time‐consuming experimental identification. In this paper the performance of Support Vector Machines (SVMs) is compared to Neural Networks and to standard statistical classification methods as Discriminant Analysis and Nearest Neighbor Classification. We show that SVMs can beat the competing methods on a dataset of 268 protein sequences to be classified into a set of 42 fold classes. We discuss misclassification with respect to biological function and similarity. In a second step we examine the performance of SVMs if the embedding is varied from frequencies of single amino acids to frequencies of tripletts of amino acids. This work shows that SVMs provide a promising alternative to standard statistical classification and prediction methods in functional genomics.     

Classification and phylogeny for the annotation of novel eukaryotic GNAT acetyltransferases

The enzymes of the GCN5-related N-acetyltransferase (GNAT) superfamily count more than 870 000 members through all kingdoms of life and share the same structural fold. GNAT enzymes transfer an acyl moiety from acyl coenzyme A to a wide range of substrates including aminoglycosides, serotonin, glucosamine-6-phosphate, protein N-termini and lysine residues of histones and other proteins. The GNAT subtype of protein N-terminal acetyltransferases (NATs) alone targets a majority of all eukaryotic proteins stressing the omnipresence of the GNAT enzymes. Despite the highly conserved GNAT fold, sequence similarity is quite low between members of this superfamily even when substrates are similar. Furthermore, this superfamily is phylogenetically not well characterized. Thus functional annotation based on sequence similarity is unreliable and strongly hampered for thousands of GNAT members that remain biochemically uncharacterized. Here we used sequence similarity networks to map the sequence space and propose a new classification for eukaryotic GNAT acetyltransferases. Using the new classification, we built a phylogenetic tree, representing the entire GNAT acetyltransferase superfamily. Our results show that protein NATs have evolved more than once on the GNAT acetylation scaffold. We use our classification to predict the function of uncharacterized sequences and verify by in vitro protein assays that two fungal genes encode NAT enzymes targeting specific protein N-terminal sequences, showing that even slight changes on the GNAT fold can lead to change in substrate specificity. In addition to providing a new map of the relationship between eukaryotic acetyltransferases the classification proposed constitutes a tool to improve functional annotation of GNAT acetyltransferases.     

Deriving topology and sequence alignment for the helix skeleton in low-resolution protein density maps

Cryoelectron microscopy (cryoEM) is an experimental technique to determine the three-dimensional (3D) structure of large protein complexes. Currently, this technique is able to generate protein density maps at 6-9 A resolution, at which the skeleton of the structure (which is composed of alpha-helices and beta-sheets) can be visualized. As a step towards predicting the entire backbone of the protein from the protein density map, we developed a method to predict the topology and sequence alignment for the skeleton helices. Our method combines the geometrical information of the skeleton helices with the Rosetta ab initio structure prediction method to derive a consensus topology and sequence alignment for the skeleton helices. We tested the method with 60 proteins. For 45 proteins, the majority of the skeleton helices were assigned a correct topology from one of our top ten predictions. The offsets of the alignment for most of the assigned helices were within +/-2 amino acids in the sequence. We also analyzed the use of the skeleton helices as a clustering tool for the decoy structures generated by Rosetta. Our comparison suggests that the topology clustering is a better method than a general overlap clustering method to enrich the ranking of decoys, particularly when the decoy pool is small.     

A polynomial-time algorithm for the matching of crossing contact-map patterns

Contact maps are a model to capture the core information in the structure of biological molecules, e.g., proteins. A contact map consists of an ordered set S of elements (representing a protein's sequence of amino acids), and a set A of element pairs of S, called arcs (representing amino acids which are closely neighbored in the structure). Given two contact maps (S, A) and (S_p, A_p ) with |A| ges |A_p| the contact map pattern matching (CMPM) problem asks whether the "pattern" (S_p, A_p) "occurs" in (S, A), i.e., informally stated, whether there is a subset of |A_p| arcs in A whose arc structure coincides with A_p . CMPM captures the biological question of finding structural motifs in protein structures. In general, CMPM is NP-hard. In this paper, we show that CMPM is solvable in O(|A|⁶|A_p| time when the pattern is {<, }-structured, i.e., when each two arcs in the pattern are disjoint or crossing. Our algorithm extends to other closely related models. In particular, it answers an open question raised by Vialette that, rephrased in terms of contact maps, asked whether CMPM for {<, } -structured patterns is NP-hard or solvable in polynomial time. Our result stands in sharp contrast to the NP-hardness of closely related problems. We provide experimental results which show that contact maps derived from real protein structures can be processed efficiently     

A protein map and its application

Graphical representation of gene sequences provides a simple way of viewing, sorting, and comparing various gene structures. Here we first report a two-dimensional graphical representation for protein sequences. With this method, we constructed the moment vectors for protein sequences, and mathematically proved that the correspondence between moment vectors and protein sequences is one-to-one. Therefore, each protein sequence can be represented as a point in a map, which we call protein map, and cluster analysis can be used for comparison between the points. Sixty-six proteins from five protein families were analyzed using this method. Our data showed that for proteins in the same family, their corresponding points in the map are close to each other. We also illustrate the efficiency of this approach by performing an extensive cluster analysis of the protein kinase C family. These results indicate that this protein map could be used to mathematically specify the similarity of two proteins and predict properties of an unknown protein based on its amino acid sequence.