A fast method for large‐scale De Novo peptide and miniprotein structure prediction

Although peptides have many biological and biomedical implications, an accurate method predicting their equilibrium structural ensembles from amino acid sequences and suitable for large‐scale experiments is still missing. We introduce a new approach—PEP‐FOLD—to the de novo prediction of peptides and miniproteins. It first predicts, in the terms of a Hidden Markov Model‐derived structural alphabet, a limited number of local conformations at each position of the structure. It then performs their assembly using a greedy procedure driven by a coarse‐grained energy score. On a benchmark of 52 peptides with 9–23 amino acids, PEP‐FOLD generates lowest‐energy conformations within 2.8 and 2.3 Å Cα root‐mean‐square deviation from the full nuclear magnetic resonance structures (NMR) and the NMR rigid cores, respectively, outperforming previous approaches. For 13 miniproteins with 27–49 amino acids, PEP‐FOLD reaches an accuracy of 3.6 and 4.6 Å Cα root‐mean‐square deviation for the most‐native and lowest‐energy conformations, using the nonflexible regions identified by NMR. PEP‐FOLD simulations are fast—a few minutes only—opening therefore, the door to in silico large‐scale rational design of new bioactive peptides and miniproteins. © 2009 Wiley Periodicals, Inc. J Comput Chem, 2010     

New energy terms for reduced protein models implemented in an off‐lattice force field

Parameterization and test calculations of a reduced protein model with new energy terms are presented. The new energy terms retain the steric properties and the most significant degrees of freedom of protein side chains in an efficient way using only one to three virtual atoms per amino acid residue. The energy terms are implemented in a force field containing predefined secondary structure elements as constraints, electrostatic interaction terms, and a solvent‐accessible surface area term to include the effect of solvation. In the force field the main‐chain peptide units are modeled as electric dipoles, which have constant directions in α‐helices and β‐sheets and variable conformation‐dependent directions in loops. Protein secondary structures can be readily modeled using these dipole terms. Parameters of the force field were derived using a large set of experimental protein structures and refined by minimizing RMS errors between the experimental structures and structures generated using molecular dynamics simulations. The final average RMS error was 3.7 Å for the main‐chain virtual atoms (C_α atoms) and 4.2 Å for all virtual atoms for a test set of 10 proteins with 58–294 amino acid residues. The force field was further tested with a substantially larger test set of 608 proteins yielding somewhat lower accuracy. The fold recognition capabilities of the force field were also evaluated using a set of 27,814 misfolded decoy structures. © 2001 John Wiley & Sons, Inc. J Comput Chem 22: 1229–1242, 2001     

SSThread: Template‐free protein structure prediction by threading pairs of contacting secondary structures followed by assembly of overlapping pairs

Acquiring the three‐dimensional structure of a protein from its amino acid sequence alone, despite a great deal of work and significant progress on the subject, is still an unsolved problem. SSThread, a new template‐free algorithm is described here that consists of making several predictions of contacting pairs of α‐helices and β‐strands derived from a database of experimental structures using a knowledge‐based potential, secondary structure prediction, and contact map prediction followed by assembly of overlapping pair predictions to create an ensemble of core structure predictions whose loops are then predicted. In a set of seven CASP10 targets SSThread outperformed the two leading methods for two targets each. The targets were all β‐strand containing structures and most of them have a high relative contact order which demonstrates the advantages of SSThread. The primary bottlenecks based on sets of 74 and 21 test cases are the pair prediction and loop prediction stages. © 2014 Wiley Periodicals, Inc.     

依据氨基酸残基的相关性预测蛋白质的结构类型

作为蛋白质的建筑构件,各种类型的蛋白质的20种氨基酸残基之间存在着特定的相互关联,反映了氨基酸残基之间的制约性,并有深刻的物理和化学的内在因素.某些氨基酸残基对之间的相关系数可以作为一种类型的蛋白质区别于其它类型蛋白质的特征,用于蛋白质结构类型的预测.研究了4种类型的蛋白质204个样品的氨基酸残基对的相关性系数,找出了可作为蛋白质结构类型特征的氨基酸残基的相关对,并用于蛋白质结构类型的预测,对于α型、β型、α/β型和α＋β型蛋白质的204个蛋白质样品的交叉测试,正确率分别为94％、89％、79％和89％,平均为88％,高于简单距离法和欧几里德距离法.     

CABS‐NMR—De novo tool for rapid global fold determination from chemical shifts,residual dipolar couplings and sparse methyl‐methyl noes

Recent development of nuclear magnetic resonance (NMR) techniques provided new types of structural restraints that can be successfully used in fast and low‐cost global protein fold determination. Here, we present CABS‐NMR, an efficient protein modeling tool, which takes advantage of such structural restraints. The restraints are converted from original NMR data to fit the coarse grained protein representation of the C‐Alpha‐Beta‐Side‐group (CABS) algorithm. CABS is a Monte Carlo search algorithm that uses a knowledge‐based force field. Its versatile structure enables a variety of protein‐modeling protocols, including purely de novo folding, folding guided by restraints derived from template structures or, structure assembly based on experimental data. In particular, CABS‐NMR uses the distance and angular restraints set derived from various NMR experiments. This new modeling technique was successfully tested in structure determination of 10 globular proteins of size up to 216 residues, for which sparse NMR data were available. Additional detailed analysis was performed for a S100A1 protein. Namely, we successfully predicted Nuclear Overhauser Effect signals on the basis of low‐energy structures obtained from chemical shifts by CABS‐NMR. It has been observed that utility of chemical shifts and other types of experimental data (i.e. residual dipolar couplings and methyl‐methyl Nuclear Overhauser Effect signals) in the presented modeling pipeline depends mainly on size of a protein and complexity of its topology. In this work, we have provided tools for either post‐experiment processing of various kinds of NMR data or fast and low‐cost structural analysis in the still challenging field of new fold predictions. © 2010 Wiley Periodicals, Inc. J Comput Chem, 2011     

A new protein graph model for function prediction

As several structural proteomic projects are producing an increasing number of protein structures with unknown function, methods that can reliably predict protein functions from protein structures are in urgent need. In this paper, we present a method to explore the clustering patterns of amino acids on the 3-dimensional space for protein function prediction. First, amino acid residues on a protein structure are clustered into spatial groups using hierarchical agglomerative clustering, based on the distance between them. Second, the protein structure is represented using a graph, where each node denotes a cluster of amino acids. The nodes are labeled with an evolutionary profile derived from the multiple alignment of homologous sequences. Then, a shortest-path graph kernel is used to calculate similarities between the graphs. Finally, a support vector machine using this graph kernel is used to train classifiers for protein function prediction. We applied the proposed method to two separate problems, namely, prediction of enzymes and prediction of DNA-binding proteins. In both cases, the results showed that the proposed method outperformed other state-of-the-art methods.     

Effective protein conformational sampling based on predicted torsion angles

Protein structure prediction is a long‐standing problem in molecular biology. Due to lack of an accurate energy function, it is often difficult to know whether the sampling algorithm or the energy function is the most important factor for failure of locating near‐native conformations of proteins. This article examines the size dependence of sampling effectiveness by using a perfect “energy function”: the root‐mean‐squared distance from the target native structure. Using protein targets up to 460 residues from critical assessment of structure prediction techniques (CASP11, 2014), we show that the accuracy of near native structures sampled is relatively independent of protein sizes but strongly depends on the errors of predicted torsion angles. Even with 40% out‐of‐range angle prediction, 2 Å or less near‐native conformation can be sampled. The result supports that the poor energy function is one of the bottlenecks of structure prediction and predicted torsion angles are useful for overcoming the bottleneck by restricting the sampling space in the absence of a perfect energy function. © 2015 Wiley Periodicals, Inc.     

Exploring the use of a structural alphabet for structural prediction of protein loops

 The prediction of loop conformations is one of the challenging problems of homology modeling, owing to the large sequence variability associated with these parts of protein structures. In the present study, we introduce a search procedure that evolves in a structural alphabet space deduced from a hidden Markov model to simplify the structural information. It uses a Bayesian criterion to predict, from the amino acid sequence of a loop region, its corresponding word in the structural alphabet space. The results show that our approach ranks 30% of the target words with the best score, 50% within the five best scores. Interestingly, our approach is also suited to accept or not the prediction performed. This allows the ranking of 57% of the target words with the best score, 67% within the five best scores, accepting 16% of learned words and rejecting 93% of unknown words. Received: 17 July 2000 / Accepted: 5 January 2001 / Published online: 3 April 2001     

Three‐dimensional structure of cyclic antibiotic teicoplanin aglycone using NMR distance and dihedral angle restraints in a DMSO solvation model

The three‐dimensional solution conformation of teicoplanin aglycone was determined using NMR spectroscopy. A combination of NOE and dihedral angle restraints in a DMSO solvation model was used to calculate an ensemble of structures having a root mean square deviation of 0.17 Å. The structures were generated using systematic searches of conformational space for optimal satisfaction of distance and dihedral angle restraints. Comparison of the NMR‐derived structure of teicoplanin aglycone with the X‐ray structure of a teicoplanin aglycone analog revealed a common backbone conformation with deviation of two aromatic side chain substituents. Experimentally determined backbone ¹³C chemical shifts showed good agreement with those computed at the density functional level of theory, providing a cross validation of the backbone conformation. The flexible portion of the molecule was consistent with the region that changes conformation to accommodate protein binding. The results showed that a hydrogen‐bonded DMSO molecule in combination with NMR‐derived restraints together enabled calculation of structures that satisfied experimental data. Copyright © 2015 John Wiley & Sons, Ltd.     

Interaction of Adjacent Amino Acids

Ramachandran plots display the dihedral angles of a single protein residue. We propose a crossed torsion angle plot called SSY‐plot between two neighboring amino acids and demonstrate that a special coherence motion can exist between some very special amino acid pairs leading to spontaneous unusual structures. A 6mer was extracted from a BBA polypeptide chain which in this plot shows two diagonal domains for the Ser‐Arg pair after some induction time. Other amino acid pairs in general do not show this kind of split domain. This shows that a special pair is required for stabilizing two distinct native structures in protein folding. We suggest that the existence of these two domains corresponds to a bifurcation between two different protein structures and that the special pair is the key to producing these two structures. These two different structures are produced spontaneously without an external agent.     

Correlations Between Amino Acids at Different Sites in Local Sequences of Protein Fragments with Given Structural Patterns

Ample evidence suggests that the local structures of peptide fragments in native proteins are to some extent encoded by their local sequences. Detecting such local correlations is important but it is still an open question what would be the most appropriate method. This is partly because conventional sequence analyses treat amino acid preferences at each site of a protein sequence independently, while it is often the inter-site interactions that bring about local sequence-structure correlations. Here a new scheme is introduced to capture the correlation between amino acid preferences at different sites for different local structure types. A library of nine-residue fragments is constructed, and the fragments are divided into clusters based on their local structures. For each local structure cluster or type, chi-square tests are used to identify correlated preferences of amino acid combinations at pairs of sites. A score function is constructed including both the single site amino acid preferences and the dual-site amino acid combination preferences, which can be used to identify whether a sequence fragment would have a strong tendency to form a particular local structure in native proteins. The results show that, given a local structure pattern, dual-site amino acid combinations contain different information from single site amino acid preferences. Representative examples show that many of the statistically identified correlations agree with previously-proposed heuristic rules about local sequence-structure correlations, or are consistent with physical-chemical interactions required to stabilize particular local structures. Results also show that such dual-site correlations in the score function significantly improves the Z-score matching a sequence fragment to its native local structure relative to nonnative local structures, and certain local structure types are highly predictable from the local sequence alone if inter-site correlations are considered.     

Contact pair dynamics during folding of two small proteins: chicken villin head piece and the Alzheimer protein beta-amyloid

The folding of an extended protein to its unique native state requires establishment of specific, predetermined, often distant, contacts between amino acid residue pairs. The dynamics of contact pair formation between various hydrophobic residues during folding of two different small proteins, the chicken villin head piece (HP-36) and the Alzheimer protein beta-amyloid (betaA-40), are investigated by Brownian dynamics (BD) simulations. These two proteins represent two very different classes-HP-36 being globular while betaA-40 is nonglobular, stringlike. Hydropathy scale and nonlocal helix propensity of amino acids are used to model the complex interaction potential among the various amino acid residues. The minimalistic model we use here employs a connected backbone chain of atoms of equal size while an amino acid is attached to each backbone atom as an additional atom of differing sizes and interaction parameters, determined by the characteristics of each amino acid. Even for such simple models, we find that the low-energy structures obtained by BD simulations of both the model proteins mimic the native state of the real protein rather well, with a best root-mean-square deviation of 4.5 A for HP-36. For betaA-40 (where a single well-defined structure is not available), the simulated structures resemble the reported ensemble rather well, with the well-known beta-bend correctly reproduced. We introduce and calculate a contact pair distance time correlation function, C(P) (ij)(t), to quantify the dynamical evolution of the pair contact formation between the amino acid residue pairs i and j. The contact pair time correlation function exhibits multistage dynamics, including a two stage fast collapse, followed by a slow (microsecond long) late stage dynamics for several specific pairs. The slow late stage dynamics is in accordance with the findings of Sali et al. Analysis of the individual trajectories shows that the slow decay is due to the attempt of the protein to form energetically more favorable pair contacts to replace the less favorable ones. This late stage contact formation is a highly cooperative process, involving participation of several pairs and thus entropically unfavorable and expected to face a large free energy barrier. This is because any new pair contact formation among hydrophobic pairs will require breaking of several contacts, before the favorable ones can be formed. This aspect of protein folding dynamics is similar to relaxation in glassy liquids, where also alpha relaxation requires highly cooperative process of hopping. The present analysis suggests that waiting time for the necessary pair contact formation may obey the Poissonian distribution. We also study the dynamics of Forster energy transfer during folding between two tagged amino acid pairs. This dynamics can be studied by fluorescence resonance energy transfer (FRET). It is found that suitably placed donor-acceptor pairs can capture the slow dynamics during folding. The dynamics probed by FRET is predicted to be nonexponential.     

Finding the needle in a haystack: educing native folds from ambiguous ab initio protein structure predictions

Current ab initio structure‐prediction methods are sometimes able to generate families of folds, one of which is native, but are unable to single out the native one due to imperfections in the folding potentials and an inability to conduct thorough explorations of the conformational space. To address this issue, here we describe a method for the detection of statistically significant folds from a pool of predicted structures. Our approach consists of clustering and averaging the structures into representative fold families. Using a metric derived from the root‐mean‐square distance (RMSD) that is less sensitive to protein size, we determine whether the simulated structures are clustered in relation to a group of random structures. The clustering method searches for cluster centers and iteratively calculates the clusters and their respective centroids. The centroid interresidue distances are adjusted by minimizing a potential constructed from the corresponding average distances of the cluster structures. Application of this method to selected proteins shows that it can detect the best fold family that is closest to native, along with several other misfolded families. We also describe a method to obtain substructures. This is useful when the folding simulation fails to give a total topology prediction but produces common subelements among the structures. We have created a web server that clusters user submitted structures, which can be found at http://bioinformatics.danforthcenter.org/services/scar. © 2001 John Wiley & Sons, Inc. J Comput Chem 22: 339–353, 2001     

pmx: Automated protein structure and topology generation for alchemical perturbations

Computational protein design requires methods to accurately estimate free energy changes in protein stability or binding upon an amino acid mutation. From the different approaches available, molecular dynamics‐based alchemical free energy calculations are unique in their accuracy and solid theoretical basis. The challenge in using these methods lies in the need to generate hybrid structures and topologies representing two physical states of a system. A custom made hybrid topology may prove useful for a particular mutation of interest, however, a high throughput mutation analysis calls for a more general approach. In this work, we present an automated procedure to generate hybrid structures and topologies for the amino acid mutations in all commonly used force fields. The described software is compatible with the Gromacs simulation package. The mutation libraries are readily supported for five force fields, namely Amber99SB, Amber99SB*‐ILDN, OPLS‐AA/L, Charmm22*, and Charmm36. © 2014 The Authors Journal of Computational Chemistry Published by Wiley Periodicals, Inc.     

Development of novel statistical potentials describing cation-pi interactions in proteins and comparison with semiempirical and quantum chemistry approaches

Novel statistical potentials derived from known protein structures are presented. They are designed to describe cation-pi and amino-pi interactions between a positively charged amino acid or an amino acid carrying a partially charged amino group and an aromatic moiety. These potentials are based on the propensity of residue types to be separated by a certain spatial distance or to have a given relative orientation. Several such potentials, describing different kinds of correlations between residue types, distances, and orientations, are derived and combined in a way that maximizes their information content and minimizes their redundancy. To test the ability of these potentials to describe cation-pi and amino-pi systems, we compare their energies with those computed with the CHARMM molecular mechanics force field and with quantum chemistry calculations at the Hartree-Fock level (HF) and at the second order of the M?ller-Plesset perturbation theory (MP2). The latter calculations are performed in the gas phase and in acetone, in order to mimic the average dielectric constant of protein environments. The energies computed with the best of our statistical potentials and with gas-phase HF or MP2 show correlation coefficients up to 0.96 when considering one side-chain degree of freedom in the statistical potentials and up to 0.94 when using a totally simplified model excluding all side-chain degrees of freedom. These potentials perform as well as, or better than, the CHARMM molecular mechanics force field that uses a much more detailed protein representation. The good performance of our cation-pi statistical potentials suggests their utility in protein structure and stability prediction and in protein design.     

Variable predictive model based classification algorithm for effective separation of protein structural classes

Variable predictive model based class discrimination (VPMCD) algorithm is proposed as an effective protein secondary structure classification tool. The algorithm mathematically represents the characteristics amino acid interactions specific to each protein structure and exploits them further to distinguish different structures. The new concept and the VPMCD classifier are established using well-studied datasets containing four protein classes as benchmark. The protein samples selected from SCOP and PDB databases with varying homology (25-100%) and non-uniform distribution of class samples provide challenging classification problem. The performance of the new method is compared with advanced classification algorithms like component coupled, SVM and neural networks. VPMCD provides superior performance for high homology datasets. 100% classification is achieved for self-consistency test and an improvement of 5% prediction accuracy is obtained during Jackknife test. The sensitivity of the new algorithm is investigated by varying model structures/types and sequence homology. Simpler to implement VPMCD algorithm is observed to be a robust classification technique and shows potential for effective extensions to other clinical diagnosis and data mining applications in biological systems.     

A new expertness index for assessment of secondary structure prediction engines

Improvement of prediction accuracy of the protein secondary structure is essential for further developments of the whole field of protein research. In this paper, the expertness of protein secondary structure prediction engines has been studied in three levels and a new criterion has been introduced in the third level. This criterion could be considered as an extension of the previous ones based on amino acid index. Using this new criterion, the expertness of some high score secondary structure prediction engines has been reanalyzed and some hidden facts have been discovered. The results of this new assessment demonstrated that a noticeable harmony has been existed among each amino acid prediction behavior in all engines. This harmony has also been seen between single global propensity and prediction accuracy of amino acid types in each secondary structure class. Moreover, it is shown that Proline and Glycine amino acids have been predicted with less accuracy in alpha helices and beta strands. In addition, regardless of different approaches used in prediction engines, beta strands have been predicted with less accuracy.