Can a physics‐based,all‐atom potential find a protein's native structure among misfolded structures? I. Large scale AMBER benchmarking

Recent work has shown that physics-based, all-atom energy functions (AMBER, CHARMM, OPLS-AA) and local minimization, when used in scoring, are able to discriminate among native and decoy structures. Yet, there have been only few instances reported of the successful use of physics based potentials in the actual refinement of protein models from a starting conformation to one that ends in structures, which are closer to the native state. An energy function that has a global minimum energy in the protein's native state and a good correlation between energy and native-likeness should be able to drive model structures closer to their native structure during a conformational search. Here, the possible reasons for the discrepancy between the scoring and refinement results for the case of AMBER potential are examined. When the conformational search via molecular dynamics is driven by the AMBER potential for a large set of 150 nonhomologous proteins and their associated decoys, often the native minimum does not appear to be the lowest free energy state. Ways of correcting the potential function in order to make it more suitable for protein model refinement are proposed.     

MOPED: method for optimizing physical energy parameters using decoys

We present a method called MOPED for optimizing energetic and structural parameters in computational models, including all-atom energy functions, when native structures and decoys are given. The present method goes beyond previous approaches in treating energy functions that are nonlinear in the parameters and continuous in the degrees of freedom. We illustrate the method by improving solvation parameters in the energy function EEF1, which consists of the CHARMM19 polar hydrogen force field augmented by a Gaussian solvation term. Although the published parameters for EEF1 correctly discriminate the native from decoys in the decoy sets of Levitt et al., they fail on several of the more difficult decoy sets of Baker et al. MOPED successfully finds improved parameters that allow EEF1 to discriminate native from decoy structures on all protein structures that do not have metals or prosthetic groups.     

Using correlated parameters for improved ranking of protein-protein docking decoys

A successful protein–protein docking study culminates in identification of decoys at top ranks with near‐native quaternary structures. However, this task remains enigmatic because no generalized scoring functions exist that effectively infer decoys according to the similarity to near‐native quaternary structures. Difficulties arise because of the highly irregular nature of the protein surface and the significant variation of the nonbonding and solvation energies based on the chemical composition of the protein–protein interface. In this work, we describe a novel method combining an interface‐size filter, a regression model for geometric compatibility (based on two correlated surface and packing parameters), and normalized interaction energy (calculated from correlated nonbonded and solvation energies), to effectively rank decoys from a set of 10,000 decoys. Tests on 30 unbound binary protein–protein complexes show that in 16 cases we can identify at least one decoy in top three ranks having ≤10 Å backbone root mean square deviation from true binding geometry. Comparisons with other state‐of‐art methods confirm the improved ranking power of our method without the use of any experiment‐guided restraints, evolutionary information, statistical propensities, or modified interaction energy equations. Tests on 118 less‐difficult bound binary protein–protein complexes with ≤35% sequence redundancy at the interface showed that in 77% cases, at least 1 in 10,000 decoys were identified with ≤5Å backbone root mean square deviation from true geometry at first rank. The work will promote the use of new concepts where correlations among parameters provide more robust scoring models. It will facilitate studies involving molecular interactions, including modeling of large macromolecular assemblies and protein structure prediction. © 2010 Wiley Periodicals, Inc. J Comput Chem, 2011.     

Inference of the solvation energy parameters of amino acids using maximum entropy approach

We present a novel technique, based on the principle of maximum entropy, for deriving the solvation energy parameters of amino acids from the knowledge of the solvent accessible areas in experimentally determined native state structures as well as high quality decoys of proteins. We present the results of detailed studies and analyze the correlations of the solvation energy parameters with the standard hydrophobic scale. We study the ability of the inferred parameters to discriminate between the native state structures of proteins and their decoy conformations.     

Assessment of programs for ligand binding affinity prediction

The prediction of the binding free energy between a ligand and a protein is an important component in the virtual screening and lead optimization of ligands for drug discovery. To determine the quality of current binding free energy estimation programs, we examined FlexX, X-Score, AutoDock, and BLEEP for their performance in binding free energy prediction in various situations including cocrystallized complex structures, cross docking of ligands to their non-cocrystallized receptors, docking of thermally unfolded receptor decoys to their ligands, and complex structures with "randomized" ligand decoys. In no case was there a satisfactory correlation between the experimental and estimated binding free energies over all the datasets tested. Meanwhile, a strong correlation between ligand molecular weight-binding affinity correlation and experimental predicted binding affinity correlation was found. Sometimes the programs also correctly ranked ligands' binding affinities even though native interactions between the ligands and their receptors were essentially lost because of receptor deformation or ligand randomization, and the programs could not decisively discriminate randomized ligand decoys from their native ligands; this suggested that the tested programs miss important components for the accurate capture of specific ligand binding interactions.     

A large decoy set of protein–protein complexes produced by flexible docking

Computational methods are needed to help characterize the structure and function of protein–protein complexes. To develop and improve such methods, standard test problems are essential. One important test is to identify experimental structures from among large sets of decoys. Here, a flexible docking procedure was used to produce such a large ensemble of decoy complexes. In addition to their use for structure prediction, they can serve as a proxy for the nonspecific, protein–protein complexes that occur transiently in the cell, which are hard to characterize experimentally, yet biochemically important. For 202 homodimers and 41 heterodimers with known X‐ray structures, we produced an average of 1217 decoys each. The structures were characterized in detail. The decoys have rather large protein–protein interfaces, with at least 45 residue–residue contacts for every 100 contacts found in the experimental complex. They have limited intramonomer deformation and limited intermonomer steric conflicts. The decoys thoroughly sample each monomer's surface, with all the surface amino acids being part of at least one decoy interface. The decoys with the lowest intramonomer deformation were analyzed separately, as proxies for nonspecific protein–protein complexes. Their interfaces are less hydrophobic than the experimental ones, with an amino acid composition similar to the overall surface composition. They have a poorer shape complementarity and a weaker association energy, but are no more fragmented than the experimental interfaces, with 2.1 distinct patches of interacting residues on average, compared to 2.6 for the experimental interfaces. The decoys should be useful for testing and parameterizing docking methods and scoring functions; they are freely available as PDB files at http://biology.polytechnique.fr/decoys . © 2010 Wiley Periodicals, Inc. J Comput Chem, 2010     

Membrane protein native state discrimination by implicit membrane models

Four implicit membrane models [IMM1, generalized Born (GB)‐surface area‐implicit membrane (GBSAIM), GB with a simple switching (GBSW), and heterogeneous dielectric GB (HDGB)] were tested for their ability to discriminate the native conformation of five membrane proteins from 450 decoys generated by the Rosetta‐Membrane program. The energy ranking of the native state and Z‐scores were used to assess the performance of the models. The effect of membrane thickness was examined and was found to be substantial. Quite satisfactory discrimination was achieved with the all‐atom IMM1 and GBSW models at 25.4 Å thickness and with the HDGB model at 28.5 Å thickness. The energy components by themselves were not discriminative. Both van der Waals and electrostatic interactions contributed to native state discrimination, to a different extent in each model. Computational efficiency of the models decreased in the order: extended‐atom IMM1 > all‐atom IMM1 > GBSAIM > GBSW > HDGB. These results encourage the further development and use of implicit membrane models for membrane protein structure prediction. © 2012 Wiley Periodicals, Inc.     

Solvent and lipid accessibility prediction as a basis for model quality assessment in soluble and membrane proteins

On-going efforts to improve protein structure prediction stimulate the development of scoring functions and methods for model quality assessment (MQA) that can be used to rank and select the best protein models for further refinement. In this work, sequence-based prediction of relative solvent accessibility (RSA) is employed as a basis for a simple MQA method for soluble proteins, and subsequently extended to the much less explored case of (alpha-helical) membrane proteins. In analogy to soluble proteins, the level of exposure to the lipid of amino acid residues in transmembrane (TM) domains is captured in terms of the relative lipid accessibility (RLA), which is predicted from sequence using low-complexity Support Vector Regression models. On an independent set of 23 TM proteins, the new SVR-based predictor yields correlation coefficient (CC) of 0.56 between the predicted and observed RLA profiles, as opposed to CC of 0.13 for a baseline predictor that utilizes TMLIP2H empirical lipophilicity scale (with standard deviations of about 0.15). A simple MQA approach is then defined by ranking models of membrane proteins in terms of consistency between predicted and observed RLA profiles, as a measure of similarity to the native structure. The new method does not require a set of decoy models to optimize parameters, circumventing current limitations in this regard. Several different sets of models, including those generated by fragment based folding simulations, and decoys obtained by swapping TM helices to mimic errors in template based assignment, are used to assess the new approach. Predicted RLA profiles can be used to successfully discriminate near native models from non-native decoys in most cases, significantly improving the separation of correct and incorrectly folded models compared to a simple baseline approach that utilizes TMLIP2H. As suggested by the robust performance of a simple MQA method for soluble proteins that utilizes more accurate RSA predictions, further significant improvements are likely to be achieved. The steady growth in the number of resolved membrane protein structures is expected to yield enhanced RLA predictions, facilitating further efforts to improve de novo and template based prediction of membrane protein structure.     

Distance dependency and minimum amino acid alphabets for decoy scoring potentials

The validity and accuracy of a proposed tertiary structure of a protein can be assessed in several ways. Scoring such a structure by a knowledge‐based potential is a well‐known approach in molecular biophysics, an important task in structure prediction and refinement, and a key step in several experiments on protein structures. Although several parameterizations for such models have been derived over the course of time, improvements in accuracy by explicitly using continuous distance information have not been suggested yet. We close this methodological gap by formulating the parameterization of a protein structure model as a linear program. Optimization of the parameters was performed using amino acid distances calculated for the residues in topology rich 2830 protein structures. We show the capability of our derived model to discriminate between native structures and decoys for a diverse set of proteins. In addition, we discuss the effect of reduced amino acid alphabets on the model. In contrast to studies focusing on binary contact schemes (without considering distance dependencies and proposing five symbols as optimal alphabet size), we find an accurate protein alphabet size to contain at least five symbols, preferably more, to assure a satisfactory fold recognition capability. © 2012 Wiley Periodicals, Inc.     

Binding energy landscape analysis helps to discriminate true hits from high-scoring decoys in virtual screening

Although virtual screening through molecular docking has been widely applied in lead discovery, it is still challenging to distinguish true hits from high-scoring decoys because of the difficulty in accurately predicting protein-ligand binding affinities. Following the successful application of energy landscape analysis to both protein folding and biomolecular binding studies, we attempted to use protein-ligand binding energy landscape analysis to recognize true binders from high-scoring decoys. Two parameters describing the binding energy landscape were used for this purpose. The energy gap, defined as the difference between the binding energy of the native binding mode and the average binding energy of other binding modes in the "denatured binding phase", was used to describe the thermodynamic stability of binding, and the number of local binding wells in the landscapes was used to account for the kinetic accessibility. These parameters, together with the docking score, were combined using logistic regression to investigate their capability to discriminate true ligands from high-scoring decoys. Inhibitors and the noninhibitors of two enzyme systems, neuraminidase and cyclooxygenase-2, were used to test their discrimination capability. Using a five-fold cross-validation, the areas under the receiver operator characteristic curves (AUCs) from the best linear combinations of parameters reached 0.878 for neuraminidase and 0.776 for cyclooxygenase-2. To make a more independent test, inhibitors and high-scoring decoys in a directory of useful decoys (DUD), the largest and most comprehensive public data set for benchmarking virtual screen programs by far, were used as independent test sets to test the discrimination capability of these parameters. The AUCs of the best linear combinations of parameters for the independent test sets were 0.750 for neuraminidase and 0.855 for cyclooxygenase-2. Furthermore, combining these two parameters with the docking scoring function improved the enrichment ratio to 200-300% compared to that using the scoring function alone. This study suggests that incorporating information from binding energy landscape analysis can significantly increase the success rate of virtual screening.     

SPICKER: a clustering approach to identify near-native protein folds

We have developed SPICKER, a simple and efficient strategy to identify near-native folds by clustering protein structures generated during computer simulations. In general, the most populated clusters tend to be closer to the native conformation than the lowest energy structures. To assess the generality of the approach, we applied SPICKER to 1489 representative benchmark proteins 相似文献   

Benchmark calculations on models of the phosphoryl transfer reaction catalyzed by protein kinase A

Protein phosphorylation has been proved to be of great importance in many stages of cell life. In the last few years, its reaction mechanism has been extensively studied. In this work we present the analysis of the performances of several computational methods with different computational costs (from multilevel to semiempirical) to point out the best method to be used at each level in the study of phosphoryl transfer. Finally, we center on the semiempirical methods, and mainly on the AM1/d Hamiltonian with different sets of parameters, which will permit hybrid quantum mechanics/molecular mechanics (QM/MM) free energy calculations on big models at an acceptable computational cost. We have used quite a large set of molecules and model reactions to test the computational methods, reproducing all the chemical steps involved in the mainly accepted reaction pathways for the protein phosphorylation. In the end, we also present the results for an enlarged model, cut out from an entire biological model: we compare the 2-D PES at the B3LYP and AM1/d levels with the purpose of obtaining a correction for the semiempirical method. The AM1/d-PhoT semiempirical parameterization corrected using single-point energy calculations at the B3LYP/MG3S level seems to be suitable to carry out reliable QM/MM calculations of the complete biological system.     

On evaluating molecular-docking methods for pose prediction and enrichment factors

Four of the most well-known, commercially available docking programs, FlexX, GOLD, GLIDE, and ICM, have been examined for their ligand-docking and virtual-screening capabilities. The relative performance of the programs in reproducing the native ligand conformation from starting SMILES strings for 164 high-resolution protein-ligand complexes is presented and compared. Applying only the native scoring functions, the latest versions of these four docking programs were also used to conduct virtual screening for 12 protein targets of therapeutic interest, involving both publicly available structures and AstraZeneca in-house structures. The capability of the four programs to correctly rank-order target-specific active compounds over alternative binders and nonbinders (decoys plus randomly selected compounds) and thereby enrich a small subset of a screening library is compared. Enrichments from the virtual-screening experiments are contrasted with those obtained with alternative 3D shape-matching and 2D similarity database-search methods.     

Application of replica exchange umbrella sampling to protein structure refinement of nontemplate models

We provide an assessment of a computational strategy for protein structure refinement that combines self‐guided Langevin dynamics with umbrella‐potential biasing replica exchange using the radius of gyration as a coordinate (R_g‐ReX). Eight structurally nonredundant proteins and their decoys were examined by sampling conformational space at room temperature using the CHARMM22/GBMV2 force field to generate the ensemble of structures. Two atomic statistical potentials (RWplus and DFIRE) were analyzed for structure identification and compared to the simulation force‐field potential. The results show that, while the R_g‐ReX simulations were able to sample conformational basins that were more structurally similar to the X‐ray crystallographic structures than the starting first‐order ranked decoys, the potentials failed to detect these basins from refinement. Of the three potential functions, RWplus yielded the highest accuracy for recognition of structures that refined to an average of nearly 20% increase in native contacts relative to the starting decoys. The overall performance of R_g‐ReX is compared to an earlier study of applying temperature‐based replica exchange to refine the same decoy sets and highlights the general challenge of achieving consistently the sampling and detection threshold of 70% fraction of native contacts. © 2013 Wiley Periodicals, Inc.     

Smooth solvation method for d-orbital semiempirical calculations of biological reactions. 2. Application to transphosphorylation thio effects in solution

Density-functional and semiempirical quantum methods and continuum dielectric and explicit solvation models are applied to study the role of solvation on the stabilization of native and thio-substituted transphosphorylation reactions. Extensive comparison is made between results obtained from the different methods. For the semiempirical methods, explicit solvation was treated using a hybrid quantum mechanical/molecular mechanical (QM/MM) approach and the implicit solvation was treated using a recently developed smooth solvation model implemented into a d-orbital semiempirical framework (MNDO/d-SCOSMO) within CHARMM. The different quantum and solvation methods were applied to the transesterification of a 3'-ribose,5'-methyl phosphodiester that serves as a nonenzymatic model for the self-cleavage reaction catalyzed by the hammerhead and hairpin ribozymes. Thio effects were studied for a double sulfur substitution at the nonbridging phosphoryl oxygen positions. The reaction profiles of both the native and double sulfur-substituted reactions from the MNDO/d-SCOSMO calculations were similar to the QM/MM results and consistent with the experimentally observed trends. These results underscore the need for a d-orbital semiempirical representation for phosphorus and sulfur for the study of experimentally observed thio effects in enzymatic and nonenzymatic phosphoryl transfer reactions. One of the major advantages of the present approach is that it can be applied to model chemical reactions at a significantly lower computational cost than either the density-functional calculations with implicit solvation or the semiempirical QM/MM simulations with explicit solvent.     

Protein model refinement using structural fragment tessellation

A method is described for the refinement of rough protein models based on finding a selection of structural fragments that match the model. Unlike most fragment-based methods, these are not necessarily contiguous in the sequence and form a tiling (tessellation) that covers most of the structure. The residue positions of the fragments are then used as a target for the model atoms to generate a revised model which is used as the basis of a subsequent pattern definition and search. The method was shown to improve the recognition of the native fold in a series of decoys largely as a result of improved secondary structure representation.     

A scoring function for docking ligands to low-resolution protein structures

We present a docking method that uses a scoring function for protein-ligand docking that is designed to maximize the docking success rate for low-resolution protein structures. We find that the resulting scoring function parameters are very different depending on whether they were optimized for high- or low-resolution protein structures. We show that this docking method can be successfully applied to predict the ligand-binding site of low-resolution structures. For a set of 25 protein-ligand complexes, in 76% of the cases, more than 50% of ligand-contacting residues are correctly predicted (using receptor crystal structures where the binding site is unspecified). Using decoys of the receptor structures having a 4 A RMSD from the native structure, for the same set of complexes, in 72% of the cases, we obtain at least one correctly predicted ligand-contacting residue. Furthermore, using an 81-protein-ligand set described by Jain, in 76 (93.8%) cases, the algorithm correctly predicts more than 50% of the ligand-contacting residues when native protein structures are used. Using 3 A RMSD from native decoys, in all but two cases (97.5%), the algorithm predicts at least one ligand-binding residue correctly. Finally, compared to the previously published Dolores method, for 298 protein-ligand pairs, the number of cases in which at least half of the specific contacts are correctly predicted is more than four times greater.     

Elastic network normal modes provide a basis for protein structure refinement

It is well recognized that thermal motions of atoms in the protein native state, the fluctuations about the minimum of the global free energy, are well reproduced by the simple elastic network models (ENMs) such as the anisotropic network model (ANM). Elastic network models represent protein dynamics as vibrations of a network of nodes (usually represented by positions of the heavy atoms or by the C(α) atoms only for coarse-grained representations) in which the spatially close nodes are connected by harmonic springs. These models provide a reliable representation of the fluctuational dynamics of proteins and RNA, and explain various conformational changes in protein structures including those important for ligand binding. In the present paper, we study the problem of protein structure refinement by analyzing thermal motions of proteins in non-native states. We represent the conformational space close to the native state by a set of decoys generated by the I-TASSER protein structure prediction server utilizing template-free modeling. The protein substates are selected by hierarchical structure clustering. The main finding is that thermal motions for some substates, overlap significantly with the deformations necessary to reach the native state. Additionally, more mobile residues yield higher overlaps with the required deformations than do the less mobile ones. These findings suggest that structural refinement of poorly resolved protein models can be significantly enhanced by reduction of the conformational space to the motions imposed by the dominant normal modes.     

Cheminformatics meets molecular mechanics: a combined application of knowledge-based pose scoring and physical force field-based hit scoring functions improves the accuracy of structure-based virtual screening

Poor performance of scoring functions is a well-known bottleneck in structure-based virtual screening (VS), which is most frequently manifested in the scoring functions' inability to discriminate between true ligands vs known nonbinders (therefore designated as binding decoys). This deficiency leads to a large number of false positive hits resulting from VS. We have hypothesized that filtering out or penalizing docking poses recognized as non-native (i.e., pose decoys) should improve the performance of VS in terms of improved identification of true binders. Using several concepts from the field of cheminformatics, we have developed a novel approach to identifying pose decoys from an ensemble of poses generated by computational docking procedures. We demonstrate that the use of target-specific pose (scoring) filter in combination with a physical force field-based scoring function (MedusaScore) leads to significant improvement of hit rates in VS studies for 12 of the 13 benchmark sets from the clustered version of the Database of Useful Decoys (DUD). This new hybrid scoring function outperforms several conventional structure-based scoring functions, including XSCORE::HMSCORE, ChemScore, PLP, and Chemgauss3, in 6 out of 13 data sets at early stage of VS (up 1% decoys of the screening database). We compare our hybrid method with several novel VS methods that were recently reported to have good performances on the same DUD data sets. We find that the retrieved ligands using our method are chemically more diverse in comparison with two ligand-based methods (FieldScreen and FLAP::LBX). We also compare our method with FLAP::RBLB, a high-performance VS method that also utilizes both the receptor and the cognate ligand structures. Interestingly, we find that the top ligands retrieved using our method are highly complementary to those retrieved using FLAP::RBLB, hinting effective directions for best VS applications. We suggest that this integrative VS approach combining cheminformatics and molecular mechanics methodologies may be applied to a broad variety of protein targets to improve the outcome of structure-based drug discovery studies.