Scoring function for automated assessment of protein structure template quality

A coarse-grained protein force field for folding and structure prediction

We have revisited the protein coarse-grained optimized potential for efficient structure prediction (OPEP). The training and validation sets consist of 13 and 16 protein targets. Because optimization depends on details of how the ensemble of decoys is sampled, trial conformations are generated by molecular dynamics, threading, greedy, and Monte Carlo simulations, or taken from publicly available databases. The OPEP parameters are varied by a genetic algorithm using a scoring function which requires that the native structure has the lowest energy, and the native-like structures have energy higher than the native structure but lower than the remote conformations. Overall, we find that OPEP correctly identifies 24 native or native-like states for 29 targets and has very similar capability to the all-atom discrete optimized protein energy model (DOPE), found recently to outperform five currently used energy models.

Localizing the membrane binding region of Group VIA Ca2+-independent phospholipase A2 using peptide amide hydrogen/deuterium exchange mass spectrometry

The Group VIA-2 Ca(2+)-independent phospholipase A(2) (GVIA-2 iPLA(2)) is composed of seven consecutive N-terminal ankyrin repeats, a linker region, and a C-terminal phospholipase catalytic domain. No structural information exists for this enzyme, and no information is known about the membrane binding surface. We carried out deuterium exchange experiments with the GVIA-2 iPLA(2) in the presence of both phospholipid substrate and the covalent inhibitor methyl arachidonoyl fluorophosphonate and located regions in the protein that change upon lipid binding. No changes were seen in the presence of only methyl arachidonoyl fluorophosphonate. The region with the greatest change upon lipid binding was region 708-730, which showed a >70% decrease in deuteration levels at numerous time points. No decreases in exchange due to phospholipid binding were seen in the ankyrin repeat domain of the protein. To locate regions with changes in exchange on the enzyme, we constructed a computational homology model based on homologous structures. This model was validated by comparing the deuterium exchange results with the predicted structure. Our model combined with the deuterium exchange results in the presence of lipid substrate have allowed us to propose the first structural model of GVIA-2 iPLA(2) as well as the interfacial lipid binding region.

Protein structure prediction by pro-Sp3-TASSER

An automated protein structure prediction algorithm, pro-sp3-Threading/ASSEmbly/Refinement (TASSER), is described and benchmarked. Structural templates are identified using five different scoring functions derived from the previously developed threading methods PROSPECTOR_3 and SP(3). Top templates identified by each scoring function are combined to derive contact and distant restraints for subsequent model refinement by short TASSER simulations. For Medium/Hard targets (those with moderate to poor quality templates and/or alignments), alternative template alignments are also generated by parametric alignment and the top models selected by TASSER-QA are included in the contact and distance restraint derivation. Then, multiple short TASSER simulations are used to generate an ensemble of full-length models. Subsequently, the top models are selected from the ensemble by TASSER-QA and used to derive TASSER contacts and distant restraints for another round of full TASSER refinement. The final models are selected from both rounds of TASSER simulations by TASSER-QA. We compare pro-sp3-TASSER with our previously developed MetaTASSER method (enhanced with chunk-TASSER for Medium/Hard targets) on a representative test data set of 723 proteins <250 residues in length. For the 348 proteins classified as easy targets (those templates with good alignments and global structure similarity to the target), the cumulative TM-score of the best of top five models by pro-sp3-TASSER shows a 2.1% improvement over MetaTASSER. For the 155/220 medium/hard targets, the improvements in TM-score are 2.8% and 2.2%, respectively. All improvements are statistically significant. More importantly, the number of foldable targets (those having models whose TM-score to native >0.4 in the top five clusters) increases from 472 to 497 for all targets, and the relative increases for medium and hard targets are 10% and 15%, respectively. A server that implements the above algorithm is available at http://cssb.biology.gatech.edu/skolnick/webservice/pro-sp3-TASSER/. The source code is also available upon request.

QMEANclust: estimation of protein model quality by combining a composite scoring function with structural density information

BACKGROUND

The selection of the most accurate protein model from a set of alternatives is a crucial step in protein structure prediction both in template-based and ab initio approaches. Scoring functions have been developed which can either return a quality estimate for a single model or derive a score from the information contained in the ensemble of models for a given sequence. Local structural features occurring more frequently in the ensemble have a greater probability of being correct. Within the context of the CASP experiment, these so called consensus methods have been shown to perform considerably better in selecting good candidate models, but tend to fail if the best models are far from the dominant structural cluster. In this paper we show that model selection can be improved if both approaches are combined by pre-filtering the models used during the calculation of the structural consensus.

RESULTS

Our recently published QMEAN composite scoring function has been improved by including an all-atom interaction potential term. The preliminary model ranking based on the new QMEAN score is used to select a subset of reliable models against which the structural consensus score is calculated. This scoring function called QMEANclust achieves a correlation coefficient of predicted quality score and GDT_TS of 0.9 averaged over the 98 CASP7 targets and perform significantly better in selecting good models from the ensemble of server models than any other groups participating in the quality estimation category of CASP7. Both scoring functions are also benchmarked on the MOULDER test set consisting of 20 target proteins each with 300 alternatives models generated by MODELLER. QMEAN outperforms all other tested scoring functions operating on individual models, while the consensus method QMEANclust only works properly on decoy sets containing a certain fraction of near-native conformations. We also present a local version of QMEAN for the per-residue estimation of model quality (QMEANlocal) and compare it to a new local consensus-based approach.

CONCLUSION

Improved model selection is obtained by using a composite scoring function operating on single models in order to enrich higher quality models which are subsequently used to calculate the structural consensus. The performance of consensus-based methods such as QMEANclust highly depends on the composition and quality of the model ensemble to be analysed. Therefore, performance estimates for consensus methods based on large meta-datasets (e.g. CASP) might overrate their applicability in more realistic modelling situations with smaller sets of models based on individual methods.

MM-align: a quick algorithm for aligning multiple-chain protein complex structures using iterative dynamic programming

Structural comparison of multiple-chain protein complexes is essential in many studies of protein-protein interactions. We develop a new algorithm, MM-align, for sequence-independent alignment of protein complex structures. The algorithm is built on a heuristic iteration of a modified Needleman-Wunsch dynamic programming (DP) algorithm, with the alignment score specified by the inter-complex residue distances. The multiple chains in each complex are first joined, in every possible order, and then simultaneously aligned with cross-chain alignments prevented. The alignments of interface residues are enhanced by an interface-specific weighting factor. MM-align is tested on a large-scale benchmark set of 205 x 3897 non-homologous multiple-chain complex pairs. Compared with a naïve extension of the monomer alignment program of TM-align, the alignment accuracy of MM-align is significantly higher as judged by the average TM-score of the physically-aligned residues. MM-align is about two times faster than TM-align because of omitting the cross-alignment zone of the DP matrix. It also shows that the enhanced alignment of the interfaces helps in identifying biologically relevant protein complex pairs.

PEP-FOLD: an online resource for de novo peptide structure prediction

Rational peptide design and large-scale prediction of peptide structure from sequence remain a challenge for chemical biologists. We present PEP-FOLD, an online service, aimed at de novo modelling of 3D conformations for peptides between 9 and 25 amino acids in aqueous solution. Using a hidden Markov model-derived structural alphabet (SA) of 27 four-residue letters, PEP-FOLD first predicts the SA letter profiles from the amino acid sequence and then assembles the predicted fragments by a greedy procedure driven by a modified version of the OPEP coarse-grained force field. Starting from an amino acid sequence, PEP-FOLD performs series of 50 simulations and returns the most representative conformations identified in terms of energy and population. Using a benchmark of 25 peptides with 9–23 amino acids, and considering the reproducibility of the runs, we find that, on average, PEP-FOLD locates lowest energy conformations differing by 2.6 Å Cα root mean square deviation from the full NMR structures. PEP-FOLD can be accessed at http://bioserv.rpbs.univ-paris-diderot.fr/PEP-FOLD

pGenTHREADER and pDomTHREADER: new methods for improved protein fold recognition and superfamily discrimination

MOTIVATION

Generation of structural models and recognition of homologous relationships for unannotated protein sequences are fundamental problems in bioinformatics. Improving the sensitivity and selectivity of methods designed for these two tasks therefore has downstream benefits for many other bioinformatics applications.

RESULTS

We describe the latest implementation of the GenTHREADER method for structure prediction on a genomic scale. The method combines profile-profile alignments with secondary-structure specific gap-penalties, classic pair- and solvation potentials using a linear combination optimized with a regression SVM model. We find this combination significantly improves both detection of useful templates and accuracy of sequence-structure alignments relative to other competitive approaches. We further present a second implementation of the protocol designed for the task of discriminating superfamilies from one another. This method, pDomTHREADER, is the first to incorporate both sequence and structural data directly in this task and improves sensitivity and selectivity over the standard version of pGenTHREADER and three other standard methods for remote homology detection.

Large-scale analysis of influenza A virus sequences reveals potential drug target sites of non-structural proteins

The non-structural protein 1 (NS1) of the influenza A virus and the NS2 protein, which is also known as nuclear export protein, play important roles in the infectious life cycle of the virus. The objective of this study was to find the degree of conservation in the NS proteins and to identify conserved sites of functional or structural importance that may be utilized as potential drug target sites. The analysis was based on 2620 amino acid sequences for the NS1 protein and 1195 sequences for the NS2 protein. The degree of conservation and potential binding sites were mapped onto the protein structures obtained from a combination of experimentally available structure fragments with predicted threading models. In addition to high conservation in protein regions of known function, novel highly conserved sites have been identified, namely Glu159, Thr171, Val192, Arg200, Glu208 and Gln218 on the NS1 protein and Ser24, Leu28, Arg66, Arg84, Ser93, Ile97 and Leu103 on the NS2 protein. Using the Q-SiteFinder binding site prediction algorithm, several highly conserved binding sites were found, including two spatially close sites on the NS1 protein, which could be targeted with a bivalent ligand that would interfere with double-stranded RNA binding. Altogether, this work reveals novel universally conserved residues that are candidates for protein-protein interactions and provide the basis for designing universal anti-influenza drugs.

Improving consensus contact prediction via server correlation reduction

Background

Protein inter-residue contacts play a crucial role in the determination and prediction of protein structures. Previous studies on contact prediction indicate that although template-based consensus methods outperform sequence-based methods on targets with typical templates, such consensus methods perform poorly on new fold targets. However, we find out that even for new fold targets, the models generated by threading programs can contain many true contacts. The challenge is how to identify them.

Results

In this paper, we develop an integer linear programming model for consensus contact prediction. In contrast to the simple majority voting method assuming that all the individual servers are equally important and independent, the newly developed method evaluates their correlation by using maximum likelihood estimation and extracts independent latent servers from them by using principal component analysis. An integer linear programming method is then applied to assign a weight to each latent server to maximize the difference between true contacts and false ones. The proposed method is tested on the CASP7 data set. If the top L/5 predicted contacts are evaluated where L is the protein size, the average accuracy is 73%, which is much higher than that of any previously reported study. Moreover, if only the 15 new fold CASP7 targets are considered, our method achieves an average accuracy of 37%, which is much better than that of the majority voting method, SVM-LOMETS, SVM-SEQ, and SAM-T06. These methods demonstrate an average accuracy of 13.0%, 10.8%, 25.8% and 21.2%, respectively.

Conclusion

Reducing server correlation and optimally combining independent latent servers show a significant improvement over the traditional consensus methods. This approach can hopefully provide a powerful tool for protein structure refinement and prediction use.

Flexible structural protein alignment by a sequence of local transformations

MOTIVATION

Throughout evolution, homologous proteins have common regions that stay semi-rigid relative to each other and other parts that vary in a more noticeable way. In order to compare the increasing number of structures in the PDB, flexible geometrical alignments are needed, that are reliable and easy to use.

RESULTS

We present a protein structure alignment method whose main feature is the ability to consider different rigid transformations at different sites, allowing for deformations beyond a global rigid transformation. The performance of the method is comparable with that of the best ones from 10 aligners tested, regarding both the quality of the alignments with respect to hand curated ones, and the classification ability. An analysis of some structure pairs from the literature that need to be matched in a flexible fashion are shown. The use of a series of local transformations can be exported to other classifiers, and a future golden protein similarity measure could benefit from it.

AVAILABILITY

A public server for the program is available at http://dmi.uib.es/ProtDeform/.

SUPPLEMENTARY INFORMATION

All data used, results and examples are available at http://dmi.uib.es/people/jairo/bio/ProtDeform.

Systematic comparison of SCOP and CATH: a new gold standard for protein structure analysis

Background

SCOP and CATH are widely used as gold standards to benchmark novel protein structure comparison methods as well as to train machine learning approaches for protein structure classification and prediction. The two hierarchies result from different protocols which may result in differing classifications of the same protein. Ignoring such differences leads to problems when being used to train or benchmark automatic structure classification methods. Here, we propose a method to compare SCOP and CATH in detail and discuss possible applications of this analysis.

Results

We create a new mapping between SCOP and CATH and define a consistent benchmark set which is shown to largely reduce errors made by structure comparison methods such as TM-Align and has useful further applications, e.g. for machine learning methods being trained for protein structure classification. Additionally, we extract additional connections in the topology of the protein fold space from the orthogonal features contained in SCOP and CATH.

Conclusion

Via an all-to-all comparison, we find that there are large and unexpected differences between SCOP and CATH w.r.t. their domain definitions as well as their hierarchic partitioning of the fold space on every level of the two classifications. A consistent mapping of SCOP and CATH can be exploited for automated structure comparison and classification.

Availability

Benchmark sets and an interactive SCOP-CATH browser are available at .

Protein structure prediction: when is it useful?

Computationally predicted three-dimensional structure of protein molecules has demonstrated the usefulness in many areas of biomedicine, ranging from approximate family assignments to precise drug screening. For nearly 40 years, however, the accuracy of the predicted models has been dictated by the availability of close structural templates. Progress has recently been achieved in refining low-resolution models closer to the native ones; this has been made possible by combining knowledge-based information from multiple sources of structural templates as well as by improving the energy funnel of physics-based force fields. Unfortunately, there has been no essential progress in the development of techniques for detecting remotely homologous templates and for predicting novel protein structures.

Cross-over between discrete and continuous protein structure space: insights into automatic classification and networks of protein structures

Structural classifications of proteins assume the existence of the fold, which is an intrinsic equivalence class of protein domains. Here, we test in which conditions such an equivalence class is compatible with objective similarity measures. We base our analysis on the transitive property of the equivalence relationship, requiring that similarity of A with B and B with C implies that A and C are also similar. Divergent gene evolution leads us to expect that the transitive property should approximately hold. However, if protein domains are a combination of recurrent short polypeptide fragments, as proposed by several authors, then similarity of partial fragments may violate the transitive property, favouring the continuous view of the protein structure space. We propose a measure to quantify the violations of the transitive property when a clustering algorithm joins elements into clusters, and we find out that such violations present a well defined and detectable cross-over point, from an approximately transitive regime at high structure similarity to a regime with large transitivity violations and large differences in length at low similarity. We argue that protein structure space is discrete and hierarchic classification is justified up to this cross-over point, whereas at lower similarities the structure space is continuous and it should be represented as a network. We have tested the qualitative behaviour of this measure, varying all the choices involved in the automatic classification procedure, i.e., domain decomposition, alignment algorithm, similarity score, and clustering algorithm, and we have found out that this behaviour is quite robust. The final classification depends on the chosen algorithms. We used the values of the clustering coefficient and the transitivity violations to select the optimal choices among those that we tested. Interestingly, this criterion also favours the agreement between automatic and expert classifications. As a domain set, we have selected a consensus set of 2,890 domains decomposed very similarly in SCOP and CATH. As an alignment algorithm, we used a global version of MAMMOTH developed in our group, which is both rapid and accurate. As a similarity measure, we used the size-normalized contact overlap, and as a clustering algorithm, we used average linkage. The resulting automatic classification at the cross-over point was more consistent than expert ones with respect to the structure similarity measure, with 86% of the clusters corresponding to subsets of either SCOP or CATH superfamilies and fewer than 5% containing domains in distinct folds according to both SCOP and CATH. Almost 15% of SCOP superfamilies and 10% of CATH superfamilies were split, consistent with the notion of fold change in protein evolution. These results were qualitatively robust for all choices that we tested, although we did not try to use alignment algorithms developed by other groups. Folds defined in SCOP and CATH would be completely joined in the regime of large transitivity violations where clustering is more arbitrary. Consistently, the agreement between SCOP and CATH at fold level was lower than their agreement with the automatic classification obtained using as a clustering algorithm, respectively, average linkage (for SCOP) or single linkage (for CATH). The networks representing significant evolutionary and structural relationships between clusters beyond the cross-over point may allow us to perform evolutionary, structural, or functional analyses beyond the limits of classification schemes. These networks and the underlying clusters are available at http://ub.cbm.uam.es/research/ProtNet.php

Making order of the fast-growing information on proteins is essential for gaining evolutionary and functional knowledge. The most successful approaches to this task are based on classifications of protein structures, such as SCOP and CATH, which assume a discrete view of the protein structure space as a collection of separated equivalence classes (folds). However, several authors proposed that protein domains should be regarded as assemblies of polypeptide fragments, which implies that the protein–structure space is continuous. Here, we assess these views of domain space through the concept of transitivity; i.e., we test whether structure similarity of A with B and B with C implies that A and C are similar, as required for consistent classification. We find that the domain space is approximately transitive and discrete at high similarity and continuous at low similarity, where transitivity is severely violated. Comparing our classification at the cross-over similarity with CATH and SCOP, we find that they join proteins at low similarity where classification is inconsistent. Part of this discrepancy is due to structural divergence of homologous domains, which are forced to be in a single cluster in CATH and SCOP. Structural and evolutionary relationships between consistent clusters are represented as a network in our approach, going beyond current protein classification schemes. We conjecture that our results are related to a change of evolutionary regime, from uniparental divergent evolution for highly related domains to assembly of large fragments for which the classical tree representation is unsuitable.

Improving consensus structure by eliminating averaging artifacts

Background

Common structural biology methods (i.e., NMR and molecular dynamics) often produce ensembles of molecular structures. Consequently, averaging of 3D coordinates of molecular structures (proteins and RNA) is a frequent approach to obtain a consensus structure that is representative of the ensemble. However, when the structures are averaged, artifacts can result in unrealistic local geometries, including unphysical bond lengths and angles.

Results

Herein, we describe a method to derive representative structures while limiting the number of artifacts. Our approach is based on a Monte Carlo simulation technique that drives a starting structure (an extended or a 'close-by' structure) towards the 'averaged structure' using a harmonic pseudo energy function. To assess the performance of the algorithm, we applied our approach to Cα models of 1364 proteins generated by the TASSER structure prediction algorithm. The average RMSD of the refined model from the native structure for the set becomes worse by a mere 0.08 Å compared to the average RMSD of the averaged structures from the native structure (3.28 Å for refined structures and 3.36 A for the averaged structures). However, the percentage of atoms involved in clashes is greatly reduced (from 63% to 1%); in fact, the majority of the refined proteins had zero clashes. Moreover, a small number (38) of refined structures resulted in lower RMSD to the native protein versus the averaged structure. Finally, compared to PULCHRA [1], our approach produces representative structure of similar RMSD quality, but with much fewer clashes.

Conclusion

The benchmarking results demonstrate that our approach for removing averaging artifacts can be very beneficial for the structural biology community. Furthermore, the same approach can be applied to almost any problem where averaging of 3D coordinates is performed. Namely, structure averaging is also commonly performed in RNA secondary prediction [2], which could also benefit from our approach.

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 .

Using least median of squares for structural superposition of flexible proteins

BACKGROUND

The conventional superposition methods use an ordinary least squares (LS) fit for structural comparison of two different conformations of the same protein. The main problem of the LS fit that it is sensitive to outliers, i.e. large displacements of the original structures superimposed.

RESULTS

To overcome this problem, we present a new algorithm to overlap two protein conformations by their atomic coordinates using a robust statistics technique: least median of squares (LMS). In order to effectively approximate the LMS optimization, the forward search technique is utilized. Our algorithm can automatically detect and superimpose the rigid core regions of two conformations with small or large displacements. In contrast, most existing superposition techniques strongly depend on the initial LS estimating for the entire atom sets of proteins. They may fail on structural superposition of two conformations with large displacements. The presented LMS fit can be considered as an alternative and complementary tool for structural superposition.

CONCLUSION

The proposed algorithm is robust and does not require any prior knowledge of the flexible regions. Furthermore, we show that the LMS fit can be extended to multiple level superposition between two conformations with several rigid domains. Our fit tool has produced successful superpositions when applied to proteins for which two conformations are known. The binary executable program for Windows platform, tested examples, and database are available from https://engineering.purdue.edu/PRECISE/LMSfit.

Fr-TM-align: a new protein structural alignment method based on fragment alignments and the TM-score

BACKGROUND

Protein tertiary structure comparisons are employed in various fields of contemporary structural biology. Most structure comparison methods involve generation of an initial seed alignment, which is extended and/or refined to provide the best structural superposition between a pair of protein structures as assessed by a structure comparison metric. One such metric, the TM-score, was recently introduced to provide a combined structure quality measure of the coordinate root mean square deviation between a pair of structures and coverage. Using the TM-score, the TM-align structure alignment algorithm was developed that was often found to have better accuracy and coverage than the most commonly used structural alignment programs; however, there were a number of situations when this was not true.

RESULTS

To further improve structure alignment quality, the Fr-TM-align algorithm has been developed where aligned fragment pairs are used to generate the initial seed alignments that are then refined using dynamic programming to maximize the TM-score. For the assessment of the structural alignment quality from Fr-TM-align in comparison to other programs such as CE and TM-align, we examined various alignment quality assessment scores such as PSI and TM-score. The assessment showed that the structural alignment quality from Fr-TM-align is better in comparison to both CE and TM-align. On average, the structural alignments generated using Fr-TM-align have a higher TM-score (~9%) and coverage (~7%) in comparison to those generated by TM-align. Fr-TM-align uses an exhaustive procedure to generate initial seed alignments. Hence, the algorithm is computationally more expensive than TM-align.

CONCLUSION

Fr-TM-align, a new algorithm that employs fragment alignment and assembly provides better structural alignments in comparison to TM-align. The source code and executables of Fr-TM-align are freely downloadable at: http://cssb.biology.gatech.edu/skolnick/files/FrTMalign/.

A novel method to map and compare protein-protein interactions in spherical viral capsids

Viral capsids are composed of multiple copies of one or a few chemically distinct capsid proteins and are mostly stabilized by inter subunit protein-protein interactions. There have been efforts to identify and analyze these protein-protein interactions, in terms of their extent and similarity, between the subunit interfaces related by quasi- and icosahedral symmetry. Here, we describe a new method to map quaternary interactions in spherical virus capsids onto polar angle space with respect to the icosahedral symmetry axes using azimuthal orthographic diagrams. This approach enables one to map the nonredundant interactions in a spherical virus capsid, irrespective of its size or triangulation number (T), onto the reference icosahedral asymmetric unit space. The resultant diagrams represent characteristic fingerprints of quaternary interactions of the respective capsids. Hence, they can be used as road maps of the protein-protein interactions to visualize the distribution and the density of the interactions. In addition, unlike the previous studies, the fingerprints of different capsids, when represented in a matrix form, can be compared with one another to quantitatively evaluate the similarity (S-score) in the subunit environments and the associated protein-protein interactions. The S-score selectively distinguishes the similarity, or lack of it, in the locations of the quaternary interactions as opposed to other well-known structural similarity metrics (e.g., RMSD, TM-score). Application of this method on a subset of T = 1 and T = 3 capsids suggests that S-score values range between 1 and 0.6 for capsids that belong to the same virus family/genus; 0.6-0.3 for capsids from different families with the same T-number and similar subunit fold; and <0.3 for comparisons of the dissimilar capsids that display different quaternary architectures (T-numbers). Finally, the sequence conserved interface residues within a virus family, whose spatial locations were also conserved have been hypothesized as the essential residues for self-assembly of the member virus capsids.

Improving accuracy of multiple sequence alignment algorithms based on alignment of neighboring residues

While most of the recent improvements in multiple sequence alignment accuracy are due to better use of vertical information, which include the incorporation of consistency-based pairwise alignments and the use of profile alignments, we observe that it is possible to further improve accuracy by taking into account alignment of neighboring residues when aligning two residues, thus making better use of horizontal information. By modifying existing multiple alignment algorithms to make use of horizontal information, we show that this strategy is able to consistently improve over existing algorithms on a few sets of benchmark alignments that are commonly used to measure alignment accuracy, and the average improvements in accuracy can be as much as 1–3% on protein sequence alignment and 5–10% on DNA/RNA sequence alignment. Unlike previous algorithms, consistent average improvements can be obtained across all identity levels.

Re-examination of structure optimization of off-lattice protein AB models by conformational space annealing

The global structural optimization is carried out for off-lattice protein AB models in two and three dimensions by conformational space annealing. The models consist of hydrophobic and hydrophilic monomers in Fibonacci sequences. To accelerate the convergence, we have introduced a shift operator in the internal coordinate system, and effectively reduced the search space by forming a quotient space. With this, we significantly improve our previous results on AB models, and provide new low energy conformations. This work provides insights on exploring complicated energy landscapes by exploiting the advantages and limitations of CSA.

Folding proteins by first-passage-times-optimized replica exchange

Replica exchange simulations have become the method of choice in computational protein science, but they still often do not allow an efficient sampling of low-energy protein configurations. Here, we reconstruct replica flow in the temperature ladder from first passage times and use it for temperature optimization, thereby maximizing sampling. The method is applied in simulations of folding thermodynamics for a number of proteins starting from the pentapeptide Met-enkephalin, through the 36-residue HP-36, up to the 67-residue protein GS-alpha3W.

Generalized pattern search algorithm for Peptide structure prediction

Finding the near-native structure of a protein is one of the most important open problems in structural biology and biological physics. The problem becomes dramatically more difficult when a given protein has no regular secondary structure or it does not show a fold similar to structures already known. This situation occurs frequently when we need to predict the tertiary structure of small molecules, called peptides. In this research work, we propose a new ab initio algorithm, the generalized pattern search algorithm, based on the well-known class of Search-and-Poll algorithms. We performed an extensive set of simulations over a well-known set of 44 peptides to investigate the robustness and reliability of the proposed algorithm, and we compared the peptide conformation with a state-of-the-art algorithm for peptide structure prediction known as PEPstr. In particular, we tested the algorithm on the instances proposed by the originators of PEPstr, to validate the proposed algorithm; the experimental results confirm that the generalized pattern search algorithm outperforms PEPstr by 21.17% in terms of average root mean-square deviation, RMSD C(alpha).

Integrated search and alignment of protein structures

MOTIVATION

Identification and comparison of similar three-dimensional (3D) protein structures has become an even greater challenge in the face of the rapidly growing structure databases. Here, we introduce Vorometric, a new method that provides efficient search and alignment of a query protein against a database of protein structures. Voronoi contacts of the protein residues are enriched with the secondary structure information and a metric substitution matrix is developed to allow efficient indexing. The contact hits obtained from a distance-based indexing method are extended to obtain high-scoring segment pairs, which are then used to generate structural alignments.

RESULTS

Vorometric is the first to address both search and alignment problems in the protein structure databases. The experimental results show that Vorometric is simultaneously effective in retrieving similar protein structures, producing high-quality structure alignments, and identifying cross-fold similarities. Vorometric outperforms current structure retrieval methods in search accuracy, while requiring com-parable running times. Furthermore, the structural superpositions produced are shown to have better quality and coverage, when compared with those of the popular structure alignment tools.

AVAILABILITY

Vorometric is available as a web service at http://bio.cse.ohio-state.edu/Vorometric

ANGLOR: a composite machine-learning algorithm for protein backbone torsion angle prediction

We developed a composite machine-learning based algorithm, called ANGLOR, to predict real-value protein backbone torsion angles from amino acid sequences. The input features of ANGLOR include sequence profiles, predicted secondary structure and solvent accessibility. In a large-scale benchmarking test, the mean absolute error (MAE) of the phi/psi prediction is 28°/46°, which is ∼10% lower than that generated by software in literature. The prediction is statistically different from a random predictor (or a purely secondary-structure-based predictor) with p-value <1.0×10⁻³⁰⁰ (or <1.0×10⁻¹⁴⁸) by Wilcoxon signed rank test. For some residues (ILE, LEU, PRO and VAL) and especially the residues in helix and buried regions, the MAE of phi angles is much smaller (10–20°) than that in other environments. Thus, although the average accuracy of the ANGLOR prediction is still low, the portion of the accurately predicted dihedral angles may be useful in assisting protein fold recognition and ab initio 3D structure modeling.

How well can the accuracy of comparative protein structure models be predicted?

Comparative structure models are available for two orders of magnitude more protein sequences than are experimentally determined structures. These models, however, suffer from two limitations that experimentally determined structures do not: They frequently contain significant errors, and their accuracy cannot be readily assessed. We have addressed the latter limitation by developing a protocol optimized specifically for predicting the Calpha root-mean-squared deviation (RMSD) and native overlap (NO3.5A) errors of a model in the absence of its native structure. In contrast to most traditional assessment scores that merely predict one model is more accurate than others, this approach quantifies the error in an absolute sense, thus helping to determine whether or not the model is suitable for intended applications. The assessment relies on a model-specific scoring function constructed by a support vector machine. This regression optimizes the weights of up to nine features, including various sequence similarity measures and statistical potentials, extracted from a tailored training set of models unique to the model being assessed: If possible, we use similarly sized models with the same fold; otherwise, we use similarly sized models with the same secondary structure composition. This protocol predicts the RMSD and NO3.5A errors for a diverse set of 580,317 comparative models of 6174 sequences with correlation coefficients (r) of 0.84 and 0.86, respectively, to the actual errors. This scoring function achieves the best correlation compared to 13 other tested assessment criteria that achieved correlations ranging from 0.35 to 0.71.

MUSTER: Improving protein sequence profile-profile alignments by using multiple sources of structure information

We develop a new threading algorithm MUSTER by extending the previous sequence profile-profile alignment method, PPA. It combines various sequence and structure information into single-body terms which can be conveniently used in dynamic programming search: (1) sequence profiles; (2) secondary structures; (3) structure fragment profiles; (4) solvent accessibility; (5) dihedral torsion angles; (6) hydrophobic scoring matrix. The balance of the weighting parameters is optimized by a grading search based on the average TM-score of 111 training proteins which shows a better performance than using the conventional optimization methods based on the PROSUP database. The algorithm is tested on 500 nonhomologous proteins independent of the training sets. After removing the homologous templates with a sequence identity to the target >30%, in 224 cases, the first template alignment has the correct topology with a TM-score >0.5. Even with a more stringent cutoff by removing the templates with a sequence identity >20% or detectable by PSI-BLAST with an E-value <0.05, MUSTER is able to identify correct folds in 137 cases with the first model of TM-score >0.5. Dependent on the homology cutoffs, the average TM-score of the first threading alignments by MUSTER is 5.1-6.3% higher than that by PPA. This improvement is statistically significant by the Wilcoxon signed rank test with a P-value < 1.0 x 10(-13), which demonstrates the effect of additional structural information on the protein fold recognition. The MUSTER server is freely available to the academic community at http://zhang.bioinformatics.ku.edu/MUSTER.

A template-finding algorithm and a comprehensive benchmark for homology modeling of proteins

The first step in homology modeling is to identify a template protein for the target sequence. The template structure is used in later phases of the calculation to construct an atomically detailed model for the target. We have built from the Protein Data Bank (PDB) a large-scale learning set that includes tens of millions of pair matches that can be either a true template or a false one. Discriminatory learning (learning from positive and negative examples) is used to train a decision tree. Each branch of the tree is a mathematical programming model. The decision tree is tested on an independent set from PDB entries and on the sequences of CASP7. It provides significant enrichment of true templates (between 50 and 100%) when compared to PSI-BLAST. The model is further verified by building atomically detailed structures for each of the tentative true templates with modeller. The probability that a true match does not yield an acceptable structural model (within 6 A RMSD from the native structure) decays linearly as a function of the TM structural-alignment score.

Protein model quality assessment prediction by combining fragment comparisons and a consensus C(alpha) contact potential

In this work, we develop a fully automated method for the quality assessment prediction of protein structural models generated by structure prediction approaches such as fold recognition servers, or ab initio methods. The approach is based on fragment comparisons and a consensus C(alpha) contact potential derived from the set of models to be assessed and was tested on CASP7 server models. The average Pearson linear correlation coefficient between predicted quality and model GDT-score per target is 0.83 for the 98 targets, which is better than those of other quality assessment methods that participated in CASP7. Our method also outperforms the other methods by about 3% as assessed by the total GDT-score of the selected top models.

Remote homology detection of integral membrane proteins using conserved sequence features

Compared with globular proteins, transmembrane proteins are surrounded by a more intricate environment and, consequently, amino acid composition varies between the different compartments. Existing algorithms for homology detection are generally developed with globular proteins in mind and may not be optimal to detect distant homology between transmembrane proteins. Here, we introduce a new profile-profile based alignment method for remote homology detection of transmembrane proteins in a hidden Markov model framework that takes advantage of the sequence constraints placed by the hydrophobic interior of the membrane. We expect that, for distant membrane protein homologs, even if the sequences have diverged too far to be recognized, the hydrophobicity pattern and the transmembrane topology are better conserved. By using this information in parallel with sequence information, we show that both sensitivity and specificity can be substantially improved for remote homology detection in two independent test sets. In addition, we show that alignment quality can be improved for the most distant homologs in a public dataset of membrane protein structures. Applying the method to the Pfam domain database, we are able to suggest new putative evolutionary relationships for a few relatively uncharacterized protein domain families, of which several are confirmed by other methods. The method is called Searcher for Homology Relationships of Integral Membrane Proteins (SHRIMP) and is available for download at http://www.sbc.su.se/shrimp/.

Contact prediction in protein modeling: scoring, folding and refinement of coarse-grained models

BACKGROUND

Several different methods for contact prediction succeeded within the Sixth Critical Assessment of Techniques for Protein Structure Prediction (CASP6). The most relevant were non-local contact predictions for targets from the most difficult categories: fold recognition-analogy and new fold. Such contacts could provide valuable structural information in case a template structure cannot be found in the PDB.

RESULTS

We described comprehensive tests of the effectiveness of contact data in various aspects of de novo modeling with CABS, an algorithm which was used successfully in CASP6 by the Kolinski-Bujnicki group. We used the predicted contacts in a simple scoring function for the post-simulation ranking of protein models and as a soft bias in the folding simulations and in the fold-refinement procedure. The latter approach turned out to be the most successful. The CABS force field used in the Replica Exchange Monte Carlo simulations cooperated with the true contacts and discriminated the false ones, which resulted in an improvement of the majority of Kolinski-Bujnicki's protein models. In the modeling we tested different sets of predicted contact data submitted to the CASP6 server. According to our results, the best performing were the contacts with the accuracy balanced with the coverage, obtained either from the best two predictors only or by a consensus from as many predictors as possible.

CONCLUSION

Our tests have shown that theoretically predicted contacts can be very beneficial for protein structure prediction. Depending on the protein modeling method, a contact data set applied should be prepared with differently balanced coverage and accuracy of predicted contacts. Namely, high coverage of contact data is important for the model ranking and high accuracy for the folding simulations.

Alternating evolutionary pressure in a genetic algorithm facilitates protein model selection

Background

Automatic protein modelling pipelines are becoming ever more accurate; this has come hand in hand with an increasingly complicated interplay between all components involved. Nevertheless, there are still potential improvements to be made in template selection, refinement and protein model selection.

Results

In the context of an automatic modelling pipeline, we analysed each step separately, revealing several non-intuitive trends and explored a new strategy for protein conformation sampling using Genetic Algorithms (GA). We apply the concept of alternating evolutionary pressure (AEP), i.e. intermediate rounds within the GA runs where unrestrained, linear growth of the model populations is allowed.

Conclusion

This approach improves the overall performance of the GA by allowing models to overcome local energy barriers. AEP enabled the selection of the best models in 40% of all targets; compared to 25% for a normal GA.

Protein model refinement using an optimized physics-based all-atom force field

One of the greatest challenges in protein structure prediction is the refinement of low-resolution predicted models to high-resolution structures that are close to the native state. Although contemporary structure prediction methods can assemble the correct topology for a large fraction of protein domains, such approximate models are often not of the resolution required for many important applications, including studies of reaction mechanisms and virtual ligand screening. Thus, the development of a method that could bring those structures closer to the native state is of great importance. We recently optimized the relative weights of the components of the Amber ff03 potential on a large set of decoy structures to create a funnel-shaped energy landscape with the native structure at the global minimum. Such an energy function might be able to drive proteins toward their native structure. In this work, for a test set of 47 proteins, with 100 decoy structures per protein that have a range of structural similarities to the native state, we demonstrate that our optimized potential can drive protein models closer to their native structure. Comparing the lowest-energy structure from each trajectory with the starting decoy, structural improvement is seen for 70% of the models on average. The ability to do such systematic structural refinements by using a physics-based all-atom potential represents a promising approach to high-resolution structure prediction.

Benchmarking of TASSER_2.0: an improved protein structure prediction algorithm with more accurate predicted contact restraints

To improve tertiary structure predictions of more difficult targets, the next generation of TASSER, TASSER_2.0, has been developed. TASSER_2.0 incorporates more accurate side-chain contact restraint predictions from a new approach, the composite-sequence method, based on consensus restraints generated by an improved threading algorithm, PROSPECTOR_3.5, which uses computationally evolved and wild-type template sequences as input. TASSER_2.0 was tested on a large-scale, benchmark set of 2591 nonhomologous, single domain proteins < or =200 residues that cover the Protein Data Bank at 35% pairwise sequence identity. Compared with the average fraction of accurately predicted side-chain contacts of 0.37 using PROSPECTOR_3.5 with wild-type template sequences, the average accuracy of the composite-sequence method increases to 0.60. The resulting TASSER_2.0 models are closer to their native structures, with an average root mean-square deviation of 4.99 A compared to the 5.31 A result of TASSER. Defining a successful prediction as a model with a root mean-square deviation to native <6.5 A, the success rate of TASSER_2.0 (TASSER) for Medium targets (targets with good templates/poor alignments) is 74.3% (64.7%) and 40.8% (35.5%) for the Hard targets (incorrect templates/alignments). For Easy targets (good templates/alignments), the success rate slightly increases from 86.3% to 88.4%.

Multiple protein sequence alignment

Multiple sequence alignments are essential in computational analysis of protein sequences and structures, with applications in structure modeling, functional site prediction, phylogenetic analysis and sequence database searching. Constructing accurate multiple alignments for divergent protein sequences remains a difficult computational task, and alignment speed becomes an issue for large sequence datasets. Here, I review methodologies and recent advances in the multiple protein sequence alignment field, with emphasis on the use of additional sequence and structural information to improve alignment quality.

Using multiple templates to improve quality of homology models in automated homology modeling

When researchers build high-quality models of protein structure from sequence homology, it is today common to use several alternative target-template alignments. Several methods can, at least in theory, utilize information from multiple templates, and many examples of improved model quality have been reported. However, to our knowledge, thus far no study has shown that automatic inclusion of multiple alignments is guaranteed to improve models without artifacts. Here, we have carried out a systematic investigation of the potential of multiple templates to improving homology model quality. We have used test sets consisting of targets from both recent CASP experiments and a larger reference set. In addition to Modeller and Nest, a new method (Pfrag) for multiple template-based modeling is used, based on the segment-matching algorithm from Levitt's SegMod program. Our results show that all programs can produce multi-template models better than any of the single-template models, but a large part of the improvement is simply due to extension of the models. Most of the remaining improved cases were produced by Modeller. The most important factor is the existence of high-quality single-sequence input alignments. Because of the existence of models that are worse than any of the top single-template models, the average model quality does not improve significantly. However, by ranking models with a model quality assessment program such as ProQ, the average quality is improved by approximately 5% in the CASP7 test set.

Progress and challenges in protein structure prediction

Depending on whether similar structures are found in the PDB library, the protein structure prediction can be categorized into template-based modeling and free modeling. Although threading is an efficient tool to detect the structural analogs, the advancements in methodology development have come to a steady state. Encouraging progress is observed in structure refinement which aims at drawing template structures closer to the native; this has been mainly driven by the use of multiple structure templates and the development of hybrid knowledge-based and physics-based force fields. For free modeling, exciting examples have been witnessed in folding small proteins to atomic resolutions. However, predicting structures for proteins larger than 150 residues still remains a challenge, with bottlenecks from both force field and conformational search.

Exploring the extremes of sequence/structure space with ensemble fold recognition in the program Phyre

Structural and functional annotation of the large and growing database of genomic sequences is a major problem in modern biology. Protein structure prediction by detecting remote homology to known structures is a well-established and successful annotation technique. However, the broad spectrum of evolutionary change that accompanies the divergence of close homologues to become remote homologues cannot easily be captured with a single algorithm. Recent advances to tackle this problem have involved the use of multiple predictive algorithms available on the Internet. Here we demonstrate how such ensembles of predictors can be designed in-house under controlled conditions and permit significant improvements in recognition by using a concept taken from protein loop energetics and applying it to the general problem of 3D clustering. We have developed a stringent test that simulates the situation where a protein sequence of interest is submitted to multiple different algorithms and not one of these algorithms can make a confident (95%) correct assignment. A method of meta-server prediction (Phyre) that exploits the benefits of a controlled environment for the component methods was implemented. At 95% precision or higher, Phyre identified 64.0% of all correct homologous query-template relationships, and 84.0% of the individual test query proteins could be accurately annotated. In comparison to the improvement that the single best fold recognition algorithm (according to training) has over PSI-Blast, this represents a 29.6% increase in the number of correct homologous query-template relationships, and a 46.2% increase in the number of accurately annotated queries. It has been well recognised in fold prediction, other bioinformatics applications, and in many other areas, that ensemble predictions generally are superior in accuracy to any of the component individual methods. However there is a paucity of information as to why the ensemble methods are superior and indeed this has never been systematically addressed in fold recognition. Here we show that the source of ensemble power stems from noise reduction in filtering out false positive matches. The results indicate greater coverage of sequence space and improved model quality, which can consequently lead to a reduction in the experimental workload of structural genomics initiatives.

QMEAN: A comprehensive scoring function for model quality assessment

In protein structure prediction, a considerable number of alternative models are usually produced from which subsequently the final model has to be selected. Thus, a scoring function for the identification of the best model within an ensemble of alternative models is a key component of most protein structure prediction pipelines. QMEAN, which stands for Qualitative Model Energy ANalysis, is a composite scoring function describing the major geometrical aspects of protein structures. Five different structural descriptors are used. The local geometry is analyzed by a new kind of torsion angle potential over three consecutive amino acids. A secondary structure-specific distance-dependent pairwise residue-level potential is used to assess long-range interactions. A solvation potential describes the burial status of the residues. Two simple terms describing the agreement of predicted and calculated secondary structure and solvent accessibility, respectively, are also included. A variety of different implementations are investigated and several approaches to combine and optimize them are discussed. QMEAN was tested on several standard decoy sets including a molecular dynamics simulation decoy set as well as on a comprehensive data set of totally 22,420 models from server predictions for the 95 targets of CASP7. In a comparison to five well-established model quality assessment programs, QMEAN shows a statistically significant improvement over nearly all quality measures describing the ability of the scoring function to identify the native structure and to discriminate good from bad models. The three-residue torsion angle potential turned out to be very effective in recognizing the native fold.

Assessing the reliability of sequence similarities detected through hydrophobic cluster analysis

Hydrophobic cluster analysis (HCA) has long been used as a tool to detect distant homologies between protein sequences, and to classify them into different folds. However, it relies on expert human intervention, and is sensitive to subjective interpretations of pattern similarities. In this study, we describe a novel algorithm to assess the similarity of hydrophobic amino acid distributions between two sequences. Our algorithm correctly identifies as misattributions several HCA-based proposals of structural similarity between unrelated proteins present in the literature. We have also used this method to identify the proper fold of a large variety of sequences, and to automatically select the most appropriate structure for homology modeling of several proteins with low sequence identity to any other member of the protein data bank. Automatic modeling of the target proteins based on these templates yielded structures with TM-scores (vs. experimental structures) above 0.60, even without further refinement. Besides enabling a reliable identification of the correct fold of an unknown sequence and the choice of suitable templates, our algorithm also shows that whereas most structural classes of proteins are very homogeneous in hydrophobic cluster composition, a tenth of the described families are compatible with a large variety of hydrophobic patterns. We have built a browsable database of every major representative hydrophobic cluster pattern present in each structural class of proteins, freely available at http://www2.ufp.pt/ pedros/HCA_db/index.htm.

A comprehensive assessment of sequence-based and template-based methods for protein contact prediction

MOTIVATION

Pair-wise residue-residue contacts in proteins can be predicted from both threading templates and sequence-based machine learning. However, most structure modeling approaches only use the template-based contact predictions in guiding the simulations; this is partly because the sequence-based contact predictions are usually considered to be less accurate than that by threading. With the rapid progress in sequence databases and machine-learning techniques, it is necessary to have a detailed and comprehensive assessment of the contact-prediction methods in different template conditions.

RESULTS

We develop two methods for protein-contact predictions: SVM-SEQ is a sequence-based machine learning approach which trains a variety of sequence-derived features on contact maps; SVM-LOMETS collects consensus contact predictions from multiple threading templates. We test both methods on the same set of 554 proteins which are categorized into 'Easy', 'Medium', 'Hard' and 'Very Hard' targets based on the evolutionary and structural distance between templates and targets. For the Easy and Medium targets, SVM-LOMETS obviously outperforms SVM-SEQ; but for the Hard and Very Hard targets, the accuracy of the SVM-SEQ predictions is higher than that of SVM-LOMETS by 12-25%. If we combine the SVM-SEQ and SVM-LOMETS predictions together, the total number of correctly predicted contacts in the Hard proteins will increase by more than 60% (or 70% for the long-range contact with a sequence separation > or =24), compared with SVM-LOMETS alone. The advantage of SVM-SEQ is also shown in the CASP7 free modeling targets where the SVM-SEQ is around four times more accurate than SVM-LOMETS in the long-range contact prediction. These data demonstrate that the state-of-the-art sequence-based contact prediction has reached a level which may be helpful in assisting tertiary structure modeling for the targets which do not have close structure templates. The maximum yield should be obtained by the combination of both sequence- and template-based predictions.

Using neural networks and evolutionary information in decoy discrimination for protein tertiary structure prediction

Background

We present a novel method of protein fold decoy discrimination using machine learning, more specifically using neural networks. Here, decoy discrimination is represented as a machine learning problem, where neural networks are used to learn the native-like features of protein structures using a set of positive and negative training examples. A set of native protein structures provides the positive training examples, while negative training examples are simulated decoy structures obtained by reversing the sequences of native structures. Various features are extracted from the training dataset of positive and negative examples and used as inputs to the neural networks.

Results

Results have shown that the best performing neural network is the one that uses input information comprising of PSI-BLAST [1] profiles of residue pairs, pairwise distance and the relative solvent accessibilities of the residues. This neural network is the best among all methods tested in discriminating the native structure from a set of decoys for all decoy datasets tested.

Conclusion

This method is demonstrated to be viable, and furthermore evolutionary information is successfully used in the neural networks to improve decoy discrimination.