Learning effective amino acid interactions through iterative stochastic techniques

Characterization on the oncogenic effect of the missense mutations of p53 via machine learning

Dysfunctions caused by missense mutations in the tumour suppressor p53 have been extensively shown to be a leading driver of many cancers. Unfortunately, it is time-consuming and labour-intensive to experimentally elucidate the effects of all possible missense variants. Recent works presented a comprehensive dataset and machine learning model to predict the functional outcome of mutations in p53. Despite the well-established dataset and precise predictions, this tool was trained on a complicated model with limited predictions on p53 mutations. In this work, we first used computational biophysical tools to investigate the functional consequences of missense mutations in p53, informing a bias of deleterious mutations with destabilizing effects. Combining these insights with experimental assays, we present two interpretable machine learning models leveraging both experimental assays and in silico biophysical measurements to accurately predict the functional consequences on p53 and validate their robustness on clinical data. Our final model based on nine features obtained comparable predictive performance with the state-of-the-art p53 specific method and outperformed other generalized, widely used predictors. Interpreting our models revealed that information on residue p53 activity, polar atom distances and changes in p53 stability were instrumental in the decisions, consistent with a bias of the properties of deleterious mutations. Our predictions have been computed for all possible missense mutations in p53, offering clinical diagnostic utility, which is crucial for patient monitoring and the development of personalized cancer treatment.

Prediction of viral-host interactions of COVID-19 by computational methods

Experimental approaches are currently used to determine viral-host interactions, but these approaches are both time-consuming and costly. For these reasons, computational-based approaches are recommended. In this study, using computational-based approaches, viral-host interactions of SARS-CoV-2 virus and human proteins were predicted. The study consists of four different stages; in the first stage viral and host protein sequences were obtained. In the second stage, protein sequences were converted into numerical expressions by various protein mapping methods. These methods are entropy-based, AVL-tree, FIBHASH, binary encoding, CPNR, PAM250, BLOSUM62, Atchley factors, Meiler parameters, EIIP, AESNN1, Miyazawa energies, Micheletti potentials, Z-scale, and hydrophobicity. In the third stage, a deep learning model was designed and BiLSTM was used for this. In the last stage, the protein sequences were classified, and the viral-host interactions were predicted. The performances of protein mapping methods were determined by accuracy, F1-score, specificity, sensitivity, and AUC scores. According to the classification results, the best classification process was obtained by the entropy-based method. With this method, 94.74% accuracy, and 0.95 AUC score were calculated. Then, the most successful classification process was performed with the Z-scale and 91.23% accuracy, and 0.96 AUC score were obtained. Although other protein mapping methods are not as efficient as Z-scale and entropy-based methods, they have achieved successful classification. AVL-tree, FIBHASH, binary encoding, CPNR, PAM250, BLOSUM62, Atchley factors, Meiler parameters and AESNN1 methods showed over 80% accuracy, F1-score, and AUC score. Accuracy scores of EIIP, Miyazawa energies, Micheletti potentials and hydrophobicity methods remained below 80%. When the results were examined in general, it was observed that the computational approaches were successful in predicting viral-host interactions between SARS-CoV-2 virus and human proteins.

Amino Acid Encoding Methods for Protein Sequences: A Comprehensive Review and Assessment

As the first step of machine-learning based protein structure and function prediction, the amino acid encoding play a fundamental role in the final success of those methods. Different from the protein sequence encoding, the amino acid encoding can be used in both residue-level and sequence-level prediction of protein properties by combining them with different algorithms. However, it has not attracted enough attention in the past decades, and there are no comprehensive reviews and assessments about encoding methods so far. In this article, we make a systematic classification and propose a comprehensive review and assessment for various amino acid encoding methods. Those methods are grouped into five categories according to their information sources and information extraction methodologies, including binary encoding, physicochemical properties encoding, evolution-based encoding, structure-based encoding, and machine-learning encoding. Then, 16 representative methods from five categories are selected and compared on protein secondary structure prediction and protein fold recognition tasks by using large-scale benchmark datasets. The results show that the evolution-based position-dependent encoding method PSSM achieved the best performance, and the structure-based and machine-learning encoding methods also show some potential for further application, the neural network based distributed representation of amino acids in particular may bring new light to this area. We hope that the review and assessment are useful for future studies in amino acid encoding.

Disordered peptide chains in an α-C-based coarse-grained model

We construct a one-bead-per-residue coarse-grained dynamical model to describe intrinsically disordered proteins at significantly longer timescales than in the all-atom models. In this model, inter-residue contacts form and disappear during the course of the time evolution. The contacts may arise between the sidechains, the backbones or the sidechains and backbones of the interacting residues. The model yields results that are consistent with many all-atom and experimental data on these systems. We demonstrate that the geometrical properties of various homopeptides differ substantially in this model. In particular, the average radius of gyration scales with the sequence length in a residue-dependent manner.

Principles and Overview of Sampling Methods for Modeling Macromolecular Structure and Dynamics

Investigation of macromolecular structure and dynamics is fundamental to understanding how macromolecules carry out their functions in the cell. Significant advances have been made toward this end in silico, with a growing number of computational methods proposed yearly to study and simulate various aspects of macromolecular structure and dynamics. This review aims to provide an overview of recent advances, focusing primarily on methods proposed for exploring the structure space of macromolecules in isolation and in assemblies for the purpose of characterizing equilibrium structure and dynamics. In addition to surveying recent applications that showcase current capabilities of computational methods, this review highlights state-of-the-art algorithmic techniques proposed to overcome challenges posed in silico by the disparate spatial and time scales accessed by dynamic macromolecules. This review is not meant to be exhaustive, as such an endeavor is impossible, but rather aims to balance breadth and depth of strategies for modeling macromolecular structure and dynamics for a broad audience of novices and experts.

Optimized distance-dependent atom-pair-based potential DOOP for protein structure prediction

The DOcking decoy-based Optimized Potential (DOOP) energy function for protein structure prediction is based on empirical distance-dependent atom-pair interactions. To optimize the atom-pair interactions, native protein structures are decomposed into polypeptide chain segments that correspond to structural motives involving complete secondary structure elements. They constitute near native ligand-receptor systems (or just pairs). Thus, a total of 8609 ligand-receptor systems were prepared from 954 selected proteins. For each of these hypothetical ligand-receptor systems, 1000 evenly sampled docking decoys with 0-10 Å interface root-mean-square-deviation (iRMSD) were generated with a method used before for protein-protein docking. A neural network-based optimization method was applied to derive the optimized energy parameters using these decoys so that the energy function mimics the funnel-like energy landscape for the interaction between these hypothetical ligand-receptor systems. Thus, our method hierarchically models the overall funnel-like energy landscape of native protein structures. The resulting energy function was tested on several commonly used decoy sets for native protein structure recognition and compared with other statistical potentials. In combination with a torsion potential term which describes the local conformational preference, the atom-pair-based potential outperforms other reported statistical energy functions in correct ranking of native protein structures for a variety of decoy sets. This is especially the case for the most challenging ROSETTA decoy set, although it does not take into account side chain orientation-dependence explicitly. The DOOP energy function for protein structure prediction, the underlying database of protein structures with hypothetical ligand-receptor systems and their decoys are freely available at http://agknapp.chemie.fu-berlin.de/doop/.

How good are simplified models for protein structure prediction?

Protein structure prediction (PSP) has been one of the most challenging problems in computational biology for several decades. The challenge is largely due to the complexity of the all-atomic details and the unknown nature of the energy function. Researchers have therefore used simplified energy models that consider interaction potentials only between the amino acid monomers in contact on discrete lattices. The restricted nature of the lattices and the energy models poses a twofold concern regarding the assessment of the models. Can a native or a very close structure be obtained when structures are mapped to lattices? Can the contact based energy models on discrete lattices guide the search towards the native structures? In this paper, we use the protein chain lattice fitting (PCLF) problem to address the first concern; we developed a constraint-based local search algorithm for the PCLF problem for cubic and face-centered cubic lattices and found very close lattice fits for the native structures. For the second concern, we use a number of techniques to sample the conformation space and find correlations between energy functions and root mean square deviation (RMSD) distance of the lattice-based structures with the native structures. Our analysis reveals weakness of several contact based energy models used that are popular in PSP.

Toward correct protein folding potentials

Empirical protein folding potentialfunctions should have a global minimum nearthe native conformationof globular proteins that fold stably, andthey should give the correct free energy offolding. We demonstrate that otherwise verysuccessful potentials fail to have even alocal minimumanywhere near the native conformation, anda seemingly well validated method ofestimatingthe thermodynamic stability of the nativestate is extremely sensitive to smallperturbations inatomic coordinates. These are bothindicative of fitting a great deal ofirrelevant detail. Here weshow how to devise a robust potentialfunction that succeeds very well at bothtasks, at least for alimited set of proteins, and this involvesdeveloping a novel representation of thedenatured state.Predicted free energies of unfolding for 25mutants of barnase are in close agreementwith theexperimental values, while for 17 mutantsthere are substantial discrepancies.

EpicCapo: epitope prediction using combined information of amino acid pairwise contact potentials and HLA-peptide contact site information

Background

Epitope identification is an essential step toward synthetic vaccine development since epitopes play an important role in activating immune response. Classical experimental approaches are laborious and time-consuming, and therefore computational methods for generating epitope candidates have been actively studied. Most of these methods, however, are based on sophisticated nonlinear techniques for achieving higher predictive performance. The use of these techniques tend to diminish their interpretability with respect to binding potential: that is, they do not provide much insight into binding mechanisms.

Results

We have developed a novel epitope prediction method named EpicCapo and its variants, EpicCapo⁺ and EpicCapo^+REF. Nonapeptides were encoded numerically using a novel peptide-encoding scheme for machine learning algorithms by utilizing 40 amino acid pairwise contact potentials (referred to as AAPPs throughout this paper). The predictive performances of EpicCapo⁺ and EpicCapo^+REF outperformed other state-of-the-art methods without losing interpretability. Interestingly, the most informative AAPPs estimated by our study were those developed by Micheletti and Simons while previous studies utilized two AAPPs developed by Miyazawa & Jernigan and Betancourt & Thirumalai. In addition, we found that all amino acid positions in nonapeptides could effect on performances of the predictive models including non-anchor positions. Finally, EpicCapo^+REF was applied to identify candidates of promiscuous epitopes. As a result, 67.1% of the predicted nonapeptides epitopes were consistent with preceding studies based on immunological experiments.

Conclusions

Our method achieved high performance in testing with benchmark datasets. In addition, our study identified a number of candidates of promiscuous CTL epitopes consistent with previously reported immunological experiments. We speculate that our techniques may be useful in the development of new vaccines. The R implementation of EpicCapo^+REF is available at http://pirun.ku.ac.th/~fsciiok/EpicCapoREF.zip. Datasets are available at http://pirun.ku.ac.th/~fsciiok/Datasets.zip.

Coarse-grain Protein Models

Coarse-graining is a powerful approach for modeling biomolecules that, over the last few decades, has been extensively applied to proteins. Coarse-grain models offer access to large systems and to slow processes without becoming computationally unmanageable. In addition, they are very versatile, enabling both the protein representation and the energy function to be adapted to the biological problem in hand. This review concentrates on modeling soluble proteins and their assemblies. It presents an overview of the coarse-grain representations, of the associated interaction potentials, and of the optimization procedures used to define them. It then shows how coarse-grain models have been used to understand processes involving proteins, from their initial folding to their functional properties, their binary interactions, and the assembly of large complexes.

Novel nonlinear knowledge-based mean force potentials based on machine learning

The prediction of 3D structures of proteins from amino acid sequences is one of the most challenging problems in molecular biology. An essential task for solving this problem with coarse-grained models is to deduce effective interaction potentials. The development and evaluation of new energy functions is critical to accurately modeling the properties of biological macromolecules. Knowledge-based mean force potentials are derived from statistical analysis of proteins of known structures. Current knowledge-based potentials are almost in the form of weighted linear sum of interaction pairs. In this study, a class of novel nonlinear knowledge-based mean force potentials is presented. The potential parameters are obtained by nonlinear classifiers, instead of relative frequencies of interaction pairs against a reference state or linear classifiers. The support vector machine is used to derive the potential parameters on data sets that contain both native structures and decoy structures. Five knowledge-based mean force Boltzmann-based or linear potentials are introduced and their corresponding nonlinear potentials are implemented. They are the DIH potential (single-body residue-level Boltzmann-based potential), the DFIRE-SCM potential (two-body residue-level Boltzmann-based potential), the FS potential (two-body atom-level Boltzmann-based potential), the HR potential (two-body residue-level linear potential), and the T32S3 potential (two-body atom-level linear potential). Experiments are performed on well-established decoy sets, including the LKF data set, the CASP7 data set, and the Decoys “R”Us data set. The evaluation metrics include the energy Z score and the ability of each potential to discriminate native structures from a set of decoy structures. Experimental results show that all nonlinear potentials significantly outperform the corresponding Boltzmann-based or linear potentials, and the proposed discriminative framework is effective in developing knowledge-based mean force potentials. The nonlinear potentials can be widely used for ab initio protein structure prediction, model quality assessment, protein docking, and other challenging problems in computational biology.

Extending the PRIME model for protein aggregation to all 20 amino acids

We extend PRIME, an intermediate-resolution protein model previously used in simulations of the aggregation of polyalanine and polyglutamine, to the description of the geometry and energetics of peptides containing all 20 amino acid residues. The 20 amino acid side chains are classified into 14 groups according to their hydrophobicity, polarity, size, charge, and potential for side chain hydrogen bonding. The parameters for extended PRIME, called PRIME 20, include hydrogen-bonding energies, side chain interaction range and energy, and excluded volume. The parameters are obtained by applying a perceptron-learning algorithm and a modified stochastic learning algorithm that optimizes the energy gap between 711 known native states from the PDB and decoy structures generated by gapless threading. The number of independent pair interaction parameters is chosen to be small enough to be physically meaningful yet large enough to give reasonably accurate results in discriminating decoys from native structures. The most physically meaningful results are obtained with 19 energy parameters.

Amino acid interaction preferences in proteins

Understanding the key factors that influence the interaction preferences of amino acids in the folding of proteins have remained a challenge. Here we present a knowledge-based approach for determining the effective interactions between amino acids based on amino acid type, their secondary structure, and the contact based environment that they find themselves in the native state structure as measured by their number of neighbors. We find that the optimal information is approximately encoded in a 60 x 60 matrix describing the 20 types of amino acids in three distinct secondary structures (helix, beta strand, and loop). We carry out a clustering scheme to understand the similarity between these interactions and to elucidate a nonredundant set. We demonstrate that the inferred energy parameters can be used for assessing the fit of a given sequence into a putative native state structure.

Four-body contact potentials derived from two protein datasets to discriminate native structures from decoys

Two-body inter-residue contact potentials for proteins have often been extracted and extensively used for threading. Here, we have developed a new scheme to derive four-body contact potentials as a way to consider protein interactions in a more cooperative model. We use several datasets of protein native structures to demonstrate that around 500 chains are sufficient to provide a good estimate of these four-body contact potentials by obtaining convergent threading results. We also have deliberately chosen two sets of protein native structures differing in resolution, one with all chains' resolution better than 1.5 A and the other with 94.2% of the structures having a resolution worse than 1.5 A to investigate whether potentials from well-refined protein datasets perform better in threading. However, potentials from well-refined proteins did not generate statistically significant better threading results. Our four-body contact potentials can discriminate well between native structures and partially unfolded or deliberately misfolded structures. Compared with another set of four-body contact potentials derived by using a Delaunay tessellation algorithm, our four-body contact potentials appear to offer a better characterization of the interactions between backbones and side chains and provide better threading results, somewhat complementary to those found using other potentials.

Scoring predictive models using a reduced representation of proteins: model and energy definition

Background

Reduced representations of proteins have been playing a keyrole in the study of protein folding. Many such models are available, with different representation detail. Although the usefulness of many such models for structural bioinformatics applications has been demonstrated in recent years, there are few intermediate resolution models endowed with an energy model capable, for instance, of detecting native or native-like structures among decoy sets. The aim of the present work is to provide a discrete empirical potential for a reduced protein model termed here PC2CA, because it employs a PseudoCovalent structure with only 2 Centers of interactions per Amino acid, suitable for protein model quality assessment.

Results

All protein structures in the set top500H have been converted in reduced form. The distribution of pseudobonds, pseudoangle, pseudodihedrals and distances between centers of interactions have been converted into potentials of mean force. A suitable reference distribution has been defined for non-bonded interactions which takes into account excluded volume effects and protein finite size. The correlation between adjacent main chain pseudodihedrals has been converted in an additional energetic term which is able to account for cooperative effects in secondary structure elements. Local energy surface exploration is performed in order to increase the robustness of the energy function.

Conclusion

The model and the energy definition proposed have been tested on all the multiple decoys' sets in the Decoys'R'us database. The energetic model is able to recognize, for almost all sets, native-like structures (RMSD less than 2.0 Å). These results and those obtained in the blind CASP7 quality assessment experiment suggest that the model compares well with scoring potentials with finer granularity and could be useful for fast exploration of conformational space. Parameters are available at the url: .

Coarse-grained protein molecular dynamics simulations

A limiting factor in biological science is the time-scale gap between experimental and computational trajectories. At this point, all-atom explicit solvent molecular dynamics (MD) are clearly too expensive to explore long-range protein motions and extract accurate thermodynamics of proteins in isolated or multimeric forms. To reach the appropriate time scale, we must then resort to coarse graining. Here we couple the coarse-grained OPEP model, which has already been used with activated methods, to MD simulations. Two test cases are studied: the stability of three proteins around their experimental structures and the aggregation mechanisms of the Alzheimer's Abeta16-22 peptides. We find that coarse-grained isolated proteins are stable at room temperature within 50 ns time scale. Based on two 220 ns trajectories starting from disordered chains, we find that four Abeta16-22 peptides can form a three-stranded beta sheet. We also demonstrate that the reptation move of one chain over the others, first observed using the activation-relaxation technique, is a kinetically important mechanism during aggregation. These results show that MD-OPEP is a particularly appropriate tool to study qualitatively the dynamics of long biological processes and the thermodynamics of molecular assemblies.

Inferring ideal amino acid interaction forms from statistical protein contact potentials

We have analyzed 29 different published matrices of protein pairwise contact potentials (CPs) between amino acids derived from different sets of proteins, either crystallographic structures taken from the Protein Data Bank (PDB) or computer-generated decoys. Each of the CPs is similar to 1 of the 2 matrices derived in the work of Miyazawa and Jernigan (Proteins 1999;34:49-68). The CP matrices of the first class can be approximated with a correlation of order 0.9 by the formula e(ij) = h(i) + h(j), 1 <or= i, j <or= 20, where the residue-type dependent factor h is highly correlated with the frequency of occurrence of a given amino acid type inside proteins. Electrostatic interactions for the potentials of this class are almost negligible. In the potentials belonging to this class, the major contribution to the potentials is the one-body transfer energy of the amino acid from water to the protein environment. Potentials belonging to the second class can be approximated with a correlation of 0.9 by the formula e(ij) = c(0) - h(i)h(j) + q(i)q(j), where c(0) is a constant, h is highly correlated with the Kyte-Doolittle hydrophobicity scale, and a new, less dominant, residue-type dependent factor q is correlated ( approximately 0.9) with amino acid isoelectric points pI. Including electrostatic interactions significantly improves the approximation for this class of potentials. While, the high correlation between potentials of the first class and the hydrophobic transfer energies is well known, the fact that this approximation can work well also for the second class of potentials is a new finding. We interpret potentials of this class as representing energies of contact of amino acid pairs within an average protein environment.

Large amplitude conformational change in proteins explored with a plastic network model: adenylate kinase

The plastic network model (PNM) is used to generate a conformational change pathway for Escherichia coli adenylate kinase based on two crystal structures, namely that of an open and a closed conformer. In this model, the energy basins corresponding to known conformers are connected at their lowest common energies. The results are used to evaluate and analyze the minimal energy pathways between these basins. The open to closed transition analysis provides an identification of hinges that is in agreement with the existing definitions based on the available X-ray structures. The elastic energy distribution and the C(alpha) pseudo-dihedral variation provide similar information on these hinges. The ensemble of the 45 published structures for this protein and closely related proteins is shown to always be within 3.0 A of the pathway, which corresponds to a conformational change between two end structures that differ by a C(alpha)-atom root-mean-squared deviation of 7.1A.

Recognizing protein folds by cluster distance geometry

Cluster distance geometry is a recent generalization of distance geometry whereby protein structures can be described at even lower levels of detail than one point per residue. With improvements in the clustering technique, protein conformations can be summarized in terms of alternative contact patterns between clusters, where each cluster contains four sequentially adjacent amino acid residues. A very simple potential function involving 210 adjustable parameters can be determined that favors the native contacts of 31 small, monomeric proteins over their respective sets of nonnative contacts. This potential then favors the native contacts for 174 small, monomeric proteins that have low sequence identity with any of the training set. A broader search finds 698 small protein chains from the Protein Data Bank where the native contacts are preferred over all alternatives, even though they have low sequence identity with the training set. This amounts to a highly predictive method for ab initio protein folding at low spatial resolution.

Geometric cooperativity and anticooperativity of three-body interactions in native proteins

Characterizing multibody interactions of hydrophobic, polar, and ionizable residues in protein is important for understanding the stability of protein structures. We introduce a geometric model for quantifying 3-body interactions in native proteins. With this model, empirical propensity values for many types of 3-body interactions can be reliably estimated from a database of native protein structures, despite the overwhelming presence of pairwise contacts. In addition, we define a nonadditive coefficient that characterizes cooperativity and anticooperativity of residue interactions in native proteins by measuring the deviation of 3-body interactions from 3 independent pairwise interactions. It compares the 3-body propensity value from what would be expected if only pairwise interactions were considered, and highlights the distinction of propensity and cooperativity of 3-body interaction. Based on the geometric model, and what can be inferred from statistical analysis of such a model, we find that hydrophobic interactions and hydrogen-bonding interactions make nonadditive contributions to protein stability, but the nonadditive nature depends on whether such interactions are located in the protein interior or on the protein surface. When located in the interior, many hydrophobic interactions such as those involving alkyl residues are anticooperative. Salt-bridge and regular hydrogen-bonding interactions, such as those involving ionizable residues and polar residues, are cooperative. When located on the protein surface, these salt-bridge and regular hydrogen-bonding interactions are anticooperative, and hydrophobic interactions involving alkyl residues become cooperative. We show with examples that incorporating 3-body interactions improves discrimination of protein native structures against decoy conformations. In addition, analysis of cooperative 3-body interaction may reveal spatial motifs that can suggest specific protein functions.

Teaching computers to fold proteins

A new general algorithm for optimization of potential functions for protein folding is introduced. It is based upon gradient optimization of the thermodynamic stability of native folds of a training set of proteins with known structure. The iterative update rule contains two thermodynamic averages which are estimated by (generalized ensemble) Monte Carlo. We test the learning algorithm on a Lennard-Jones (LJ) force field with a torsional angle degrees-of-freedom and a single-atom side-chain. In a test with 24 peptides of known structure, none folded correctly with the initial potential functions, but two-thirds came within 3 A to their native fold after optimizing the potential functions.

Accurate and efficient description of protein vibrational dynamics: Comparing molecular dynamics and Gaussian models

Current all-atom potential based molecular dynamics (MD) allows the identification of a protein's functional motions on a wide-range of timescales, up to few tens of nanoseconds. However, functional, large-scale motions of proteins may occur on a timescale currently not accessible by all-atom potential based MD. To avoid the massive computational effort required by this approach, several simplified schemes have been introduced. One of the most satisfactory is the Gaussian network approach based on the energy expansion in terms of the deviation of the protein backbone from its native configuration. Here, we consider an extension of this model that captures in a more realistic way the distribution of native interactions due to the introduction of effective side-chain centroids. Since their location is entirely determined by the protein backbone, the model is amenable to the same exact and computationally efficient treatment as previous simpler models. The ability of the model to describe the correlated motion of protein residues in thermodynamic equilibrium is established through a series of successful comparisons with an extensive (14 ns) MD simulation based on the AMBER potential of HIV-1 protease in complex with a peptide substrate. Thus, the model presented here emerges as a powerful tool to provide preliminary, fast yet accurate characterizations of protein near-native motion.

Can Contact Potentials Reliably Predict Stability of Proteins?

The simplest approximation of interaction potential between amino acid residues in proteins is the contact potential, which defines the effective free energy of a protein conformation by a set of amino acid contacts formed in this conformation. Finding a contact potential capable of predicting free energies of protein states across a variety of protein families will aid protein folding and engineering in silico on a computationally tractable time-scale. We test the ability of contact potentials to accurately and transferably (across various protein families) predict stability changes of proteins upon mutations. We develop a new methodology to determine the contact potentials in proteins from experimental measurements of changes in protein's thermodynamic stabilities (DeltaDeltaG) upon mutations. We apply our methodology to derive sets of contact interaction parameters for a hierarchy of interaction models including solvation and multi-body contact parameters. We test how well our models reproduce experimental measurements by statistical tests. We evaluate the maximum accuracy of predictions obtained by using contact potentials and the correlation between parameters derived from different data-sets of experimental (DeltaDeltaG) values. We argue that it is impossible to reach experimental accuracy and derive fully transferable contact parameters using the contact models of potentials. However, contact parameters may yield reliable predictions of DeltaDeltaG for datasets of mutations confined to the same amino acid positions in the sequence of a single protein.

Elucidation of the disulfide-folding pathway of hirudin by a topology-based approach

A theoretical model for the folding of proteins containing disulfide bonds is introduced. The model exploits the knowledge of the native state to favor the progressive establishment of native interactions. At variance with traditional approaches based on native topology, not all native bonds are treated in the same way; in particular, a suitable energy term is introduced to account for the special strength of disulfide bonds, as well as their ability to undergo intramolecular reshuffling. The model thus possesses the minimal ingredients necessary to investigate the much debated issue of whether the refolding process occurs through partially structured intermediates with native or non-native disulfide bonds. This strategy is applied to a context of particular interest, the refolding process of hirudin, a thrombin-specific protease inhibitor, for which conflicting folding pathways have been proposed. We show that the only two parameters in the model (temperature and disulfide strength) can be tuned to reproduce well a set of experimental transitions between species with different number of formed disulfides. This model is then used to provide a characterization of the folding process and a detailed description of the species involved in the rate-limiting step of hirudin refolding.

Simplicial edge representation of protein structures and alpha contact potential with confidence measure

Protein representation and potential function are two important ingredients for studying protein folding, equilibrium thermodynamics, and sequence design. We introduce a novel geometric representation of protein contact interactions using the edge simplices from the alpha shape of the protein structure. This representation can eliminate implausible neighbors that are not in physical contact, and can avoid spurious contact between two residues when a third residue is between them. We developed statistical alpha contact potential using an odds-ratio model. A studentized bootstrap method was then introduced to assess the 95% confidence intervals for each of the 210 propensity parameters. We found, with confidence, that there is significant long-range propensity (>30 residues apart) for hydrophobic interactions. We tested alpha contact potential for native structure discrimination using several sets of decoy structures, and found that it often performs comparably with atom-based potentials requiring many more parameters. We also show that accurate geometric representation is important, and that alpha contact potential has better performance than potential defined by cutoff distance between geometric centers of side chains. Hierarchical clustering of alpha contact potentials reveals natural grouping of residues. To explore the relationship between shape and physicochemical representations, we tested the minimum alphabet size necessary for native structure discrimination. We found that there is no significant difference in performance of discrimination when alphabet size varies from 7 to 20, if geometry is represented accurately by alpha simplicial edges. This result suggests that the geometry of packing plays an important role, but the specific residue types are often interchangeable.

Assembly of protein tertiary structures from secondary structures using optimized potentials

We present a simulated annealing-based method for the prediction of the tertiary structures of proteins given knowledge of the secondary structure associated with each amino acid in the sequence. The backbone is represented in a detailed fashion whereas the sidechains and pairwise interactions are modeled in a simplified way, following the LINUS model of Srinivasan and Rose. A perceptron-based technique is used to optimize the interaction potentials for a training set of three proteins. For these proteins, the procedure is able to reproduce the tertiary structures to below 3 A in root mean square deviation (rmsd) from the PDB targets. We present the results of tests on twelve other proteins. For half of these, the lowest energy decoy has a rmsd from the native state below 6 A and, in 9 out of 12 cases, we obtain decoys whose rmsd from the native states are also well below 5 A.

Optimal potentials for predicting inter-helical packing in transmembrane proteins

A set of pairwise contact potentials between amino acid residues in transmembrane helices was determined from the known native structure of the transmembrane protein (TMP) bacteriorhodopsin by the method of perceptron learning, using Monte Carlo dynamics to generate suitable "decoy" structures. The procedure of finding these decoys is simpler than for globular proteins, since it is reasonable to assume that helices behave as independent, stable objects and, therefore, the search in the conformational space is greatly reduced. With the learnt potentials, the association of the helices in bacteriorhodopsin was successfully simulated. The folding of a second TMP (the helix-dimer glycophorin A) was then accomplished with only a refinement of the potentials from a small number of decoys.

A protein folding potential that places the native states of a large number of proteins near a local minimum

BACKGROUND

We present a simple method to train a potential function for the protein folding problem which, even though trained using a small number of proteins, is able to place a significantly large number of native conformations near a local minimum. The training relies on generating decoys by energy minimization of the native conformations using the current potential and using a physically meaningful objective function (derivative of energy with respect to torsion angles at the native conformation) during the quadratic programming to place the native conformation near a local minimum.

RESULTS

We also compare the performance of three different types of energy functions and find that while the pairwise energy function is trainable, a solvation energy function by itself is untrainable if decoys are generated by minimizing the current potential starting at the native conformation. The best results are obtained when a pairwise interaction energy function is used with solvation energy function.

CONCLUSIONS

We are able to train a potential function using six proteins which places a total of 42 native conformations within approximately 4 A rmsd and 71 native conformations within approximately 6 A rmsd of a local minimum out of a total of 91 proteins. Furthermore, the threading test using the same 91 proteins ranks 89 native conformations to be first and the other two as second.

Crucial stages of protein folding through a solvable model: predicting target sites for enzyme-inhibiting drugs

An exactly solvable model based on the topology of a protein native state is applied to identify bottlenecks and key sites for the folding of human immunodeficiency virus type 1 (HIV-1) protease. The predicted sites are found to correlate well with clinical data on resistance to Food and Drug Administration-approved drugs. It has been observed that the effects of drug therapy are to induce multiple mutations on the protease. The sites where such mutations occur correlate well with those involved in folding bottlenecks identified through the deterministic procedure proposed in this study. The high statistical significance of the observed correlations suggests that the approach may be promisingly used in conjunction with traditional techniques to identify candidate locations for drug attacks.

Folding funnels: the key to robust protein structure prediction

Natural proteins fold because their free energy landscapes are funneled to their native states. The degree to which a model energy function for protein structure prediction can avoid the multiple minima problem and reliably yield at least low-resolution predictions is also dependent on the topography of the energy landscape. We show that the degree of funneling can be quantitatively expressed in terms of a few averaged properties of the landscape. This allows us to optimize simplified energy functions for protein structure prediction even in the absence of homology information. Here we outline the optimization procedure in the context of associative memory energy functions originally introduced for tertiary structure recognition and demonstrate that even partially funneled landscapes lead to qualitatively correct, low-resolution predictions.

How to guarantee optimal stability for most representative structures in the Protein Data Bank

We proposed recently an optimization method to derive energy parameters for simplified models of protein folding. The method is based on the maximization of the thermodynamic average of the overlap between protein native structures and a Boltzmann ensemble of alternative structures. Such a condition enforces protein models whose ground states are most similar to the corresponding native states. We present here an extensive testing of the method for a simple residue-residue contact energy function and for alternative structures generated by threading. The optimized energy function guarantees high stability and a well-correlated energy landscape to most representative structures in the PDB database. Failures in the recognition of the native structure can be attributed to the neglect of interactions between different chains in oligomeric proteins or with cofactors. When these are taken into account, only very few X-ray structures are not recognized. Most of them are short inhibitors or fragments and one is a structure that presents serious inconsistencies. Finally, we discuss the reasons that make NMR structures more difficult to recognizeCopyright 2001 Wiley-Liss, Inc.