Accurate Prediction Of Binding Affinity Research Articles

The Effect of Conformational Flexibility on Binding Free Energy Estimation between Kinases and Their Inhibitors.

Accurate prediction of binding affinities of drug candidates to their targets remains challenging because of protein flexibility in solution. Conformational flexibility of the ATP-binding site in the CDK2 and ERK2 kinases was identified using molecular dynamics simulations. The binding free energy (ΔG) of twenty-four ATP-competitive inhibitors toward these kinases was assessed using an alchemical free energy perturbation method, MP-CAFEE. However, large calculation errors of 2-3 kcal/mol were observed using this method, where the free energy simulation starts from a single equilibrated conformation. Here, we developed a new ΔG computation method, where the starting structure was set to multiconformations to cover flexibility. The calculation accuracy was successfully improved, especially for larger molecular size compounds, leading to reliable prediction of a broader range of drug candidates. The present study demonstrates that conformational flexibility of interactions between a compound and the glycine-rich loop in the kinases is a key factor in ΔG estimation.

Accurate calculation of the absolute free energy of binding for drug molecules.

Accurate prediction of binding affinities has been a central goal of computational chemistry for decades, yet remains elusive. Despite good progress, the required accuracy for use in a drug-discovery context has not been consistently achieved for drug-like molecules. Here, we perform absolute free energy calculations based on a thermodynamic cycle for a set of diverse inhibitors binding to bromodomain-containing protein 4 (BRD4) and demonstrate that a mean absolute error of 0.6 kcal mol-1 can be achieved. We also show a similar level of accuracy (1.0 kcal mol-1) can be achieved in pseudo prospective approach. Bromodomains are epigenetic mark readers that recognize acetylation motifs and regulate gene transcription, and are currently being investigated as therapeutic targets for cancer and inflammation. The unprecedented accuracy offers the exciting prospect that the binding free energy of drug-like compounds can be predicted for pharmacologically relevant targets.

Evaluation of GalaxyDock Based on the Community Structure-Activity Resource 2013 and 2014 Benchmark Studies.

We analyze the results of the GalaxyDock protein-ligand docking program in the two recent experiments of Community Structure-Activity Resource (CSAR), CSAR 2013 and 2014. GalaxyDock performs global optimization of a modified AutoDock3 energy function by employing the conformational space annealing method. The energy function of GalaxyDock is quite sensitive to atomic clashes. Such energy functions can be effective for sampling physically correct conformations but may not be effective for scoring when conformations are not fully optimized. In phase 1 of CSAR 2013, we successfully selected all four true binders of digoxigenin along with three false positives. However, the energy values were rather high due to insufficient optimization of the conformations docked to homology models. A posteriori relaxation of the model complex structures by GalaxyRefine improved the docking energy values and differentiated the true binders from the false positives better. In the scoring test of CSAR 2013 phase 2, we selected the best poses for each of the two targets. The results of CSAR 2013 phase 3 suggested that an improved method for generating initial conformations for GalaxyDock is necessary for targets involving bulky ligands. Finally, combining existing binding information with GalaxyDock energy-based optimization may be needed for more accurate binding affinity prediction.

The Augmenting Effects of Desolvation and Conformational Energy Terms on the Predictions of Docking Programs against mPGES-1.

In this study we introduce a rescoring method to improve the accuracy of docking programs against mPGES-1. The rescoring method developed is a result of extensive computational study in which different scoring functions and molecular descriptors were combined to develop consensus and rescoring methods. 127 mPGES-1 inhibitors were collected from literature and were segregated into training and external test sets. Docking of the 27 training set compounds was carried out using default settings in AutoDock Vina, AutoDock, DOCK6 and GOLD programs. The programs showed low to moderate correlation with the experimental activities. In order to introduce the contributions of desolvation penalty and conformation energy of the inhibitors various molecular descriptors were calculated. Later, rescoring method was developed as empirical sum of normalised values of docking scores, LogP and Nrotb. The results clearly indicated that LogP and Nrotb recuperate the predictions of these docking programs. Further the efficiency of the rescoring method was validated using 100 test set compounds. The accurate prediction of binding affinities for analogues of the same compounds is a major challenge for many of the existing docking programs; in the present study the high correlation obtained for experimental and predicted pIC50 values for the test set compounds validates the efficiency of the scoring method.

BgN-Score and BsN-Score: bagging and boosting based ensemble neural networks scoring functions for accurate binding affinity prediction of protein-ligand complexes.

BackgroundAccurately predicting the binding affinities of large sets of protein-ligand complexes is a key challenge in computational biomolecular science, with applications in drug discovery, chemical biology, and structural biology. Since a scoring function (SF) is used to score, rank, and identify drug leads, the fidelity with which it predicts the affinity of a ligand candidate for a protein's binding site has a significant bearing on the accuracy of virtual screening. Despite intense efforts in developing conventional SFs, which are either force-field based, knowledge-based, or empirical, their limited predictive power has been a major roadblock toward cost-effective drug discovery. Therefore, in this work, we present novel SFs employing a large ensemble of neural networks (NN) in conjunction with a diverse set of physicochemical and geometrical features characterizing protein-ligand complexes to predict binding affinity.ResultsWe assess the scoring accuracies of two new ensemble NN SFs based on bagging (BgN-Score) and boosting (BsN-Score), as well as those of conventional SFs in the context of the 2007 PDBbind benchmark that encompasses a diverse set of high-quality protein families. We find that BgN-Score and BsN-Score have more than 25% better Pearson's correlation coefficient (0.804 and 0.816 vs. 0.644) between predicted and measured binding affinities compared to that achieved by a state-of-the-art conventional SF. In addition, these ensemble NN SFs are also at least 19% more accurate (0.804 and 0.816 vs. 0.675) than SFs based on a single neural network that has been traditionally used in drug discovery applications. We further find that ensemble models based on NNs surpass SFs based on the decision-tree ensemble technique Random Forests.ConclusionsEnsemble neural networks SFs, BgN-Score and BsN-Score, are the most accurate in predicting binding affinity of protein-ligand complexes among the considered SFs. Moreover, their accuracies are even higher when they are used to predict binding affinities of protein-ligand complexes that are related to their training sets.

Does a More Precise Chemical Description of Protein–Ligand Complexes Lead to More Accurate Prediction of Binding Affinity?

Predicting the binding affinities of large sets of diverse molecules against a range of macromolecular targets is an extremely challenging task. The scoring functions that attempt such computational prediction are essential for exploiting and analyzing the outputs of docking, which is in turn an important tool in problems such as structure-based drug design. Classical scoring functions assume a predetermined theory-inspired functional form for the relationship between the variables that describe an experimentally determined or modeled structure of a protein–ligand complex and its binding affinity. The inherent problem of this approach is in the difficulty of explicitly modeling the various contributions of intermolecular interactions to binding affinity. New scoring functions based on machine-learning regression models, which are able to exploit effectively much larger amounts of experimental data and circumvent the need for a predetermined functional form, have already been shown to outperform a broad range of state-of-the-art scoring functions in a widely used benchmark. Here, we investigate the impact of the chemical description of the complex on the predictive power of the resulting scoring function using a systematic battery of numerical experiments. The latter resulted in the most accurate scoring function to date on the benchmark. Strikingly, we also found that a more precise chemical description of the protein–ligand complex does not generally lead to a more accurate prediction of binding affinity. We discuss four factors that may contribute to this result: modeling assumptions, codependence of representation and regression, data restricted to the bound state, and conformational heterogeneity in data.

Prediction of SAMPL4 host–guest binding affinities using funnel metadynamics

Accurately predicting binding affinities between ligands and macromolecules has been a much sought-after goal. A tremendous amount of resources can be saved in the pharmaceutical industry through accurate binding-affinity prediction and hence correct decision-making for the drug discovery processes. Owing to the structural complexity of macromolecules, one of the issues in binding affinity prediction using molecular dynamics is the adequate sampling of the conformational space. Recently, the funnel metadynamics method (Limongelli et al. in Proc Natl Acad Sci USA 110:6358, 2013) was developed to enhance the sampling of the ligand at the binding site as well as in the solvated state, and offer the possibility to predict the absolute binding free energy. We apply funnel metadynamics to predict host-guest binding affinities for the cucurbit[7]uril host as part of the SAMPL4 blind challenge. Using total simulation times of 300-400 ns per ligand, we show that the errors due to inadequate sampling are below 1 kcal/mol. However, despite the large investment in terms of computational time, the results compared to experiment are not better than a random guess. As we obtain differences of up to 11 kcal/mol when switching between two commonly used force fields (with automatically generated parameters), we strongly believe that in the pursuit of accurate binding free energies a more careful force-field parametrization is needed to address this type of system.

Multipose Binding in Molecular Docking

Molecular docking has been extensively applied in virtual screening of small molecule libraries for lead identification and optimization. A necessary prerequisite for successful differentiation between active and non-active ligands is the accurate prediction of their binding affinities in the complex by use of docking scoring functions. However, many studies have shown rather poor correlations between docking scores and experimental binding affinities. Our work aimed to improve this correlation by implementing a multipose binding concept in the docking scoring scheme. Multipose binding, i.e., the property of certain protein-ligand complexes to exhibit different ligand binding modes, has been shown to occur in nature for a variety of molecules. We conducted a high-throughput docking study and implemented multipose binding in the scoring procedure by considering multiple docking solutions in binding affinity prediction. In general, improvement of the agreement between docking scores and experimental data was observed, and this was most pronounced in complexes with large and flexible ligands and high binding affinities. Further developments of the selection criteria for docking solutions for each individual complex are still necessary for a general utilization of the multipose binding concept for accurate binding affinity prediction by molecular docking.

Postprocessing of molecular docking poses using binding free energy calculations

The main key to success in structure-based drug discovery is the accurate prediction of binding affinities of hit compounds. Molecular docking and scoring functions are often used for this purpose. However, it is often found that the top-ranked docking poses do not represent the right binding mode, and frequently there is no correlation between docking score and biological data. Therefore, “post-processing” of docking poses has recently got attraction. In previous work [1], we have successfully computed binding free energies (MM/PBSA) of 222 Wee1 kinase inhibitors and used the derived validated models for virtual screening. In the current work, we extended our studies to a data set of PIM1 kinase inhibitors. Cross-docking studies showed that the correct binding mode of the inhibitors can be determined after applying a post-processing procedure. The top-ranked docking poses gave wrong binding mode (high RMSD values ~4.0 A), whereas the top-ranked poses selected after postprocessing yielded RMSD values around 0.5 A. Subsequently, the docking poses giving the lowest binding free energy were selected and these values were used to establish a correlation with experimental data. A significant correlation between ΔGcal. and ΔGexp (r2 = 0.58) was obtained. To summarize, the protocol described in the current work can be used for postprocessing of protein-ligand docking poses and for predicting biological activities of novel hits. Therefore, the protocol can be applied for structure-based optimization of hit molecules.

Fast Estimation of Solvation Free Energies for Diverse Chemical Species

The free energy of solvation can play an important or even dominant role in the accurate prediction of binding affinities and various other molecular-scale interaction phenomena critical to the study of biochemical processes. Many research applications for solvation modeling, such as fragment-based drug design, require algorithms that are both accurate and computationally inexpensive. We have developed a calculation of solvation free energy which runs fast enough for interactive applications, functions for a wide range of chemical species relevant to simulating molecules for biological and pharmaceutical applications, and is readily extended when data for new species becomes available. We have also demonstrated that the incorporation of ab initio data provides necessary access to sufficient reference data for a broad range of chemical features. Our empirical model, including an electrostatic term and a different set of atom types, demonstrates improvements over a previous, solvent-accessible surface area-only model by Wang et al. when fit to identical training sets (mean absolute error of 0.513 kcal/mol versus the 0.538 kcal/mol reported by Wang). The incorporation of ab initio solvation free energies provides a significant increase in the breadth of chemical features for which the model can be applied by introducing classes of compounds for which little or no experimental data is available. The increased breadth and the speed of this solvation model allow for conformational minimization, conformational search, and ligand binding free energy calculations that economically account for the complex interplay of bonded, nonbonded, and solvation free energies as conformations with varying solvent-accessible surfaces are sampled.

CSCORE: A SIMPLE YET EFFECTIVE SCORING FUNCTION FOR PROTEIN–LIGAND BINDING AFFINITY PREDICTION USING MODIFIED CMAC LEARNING ARCHITECTURE

Protein-ligand docking is a computational method to identify the binding mode of a ligand and a target protein, and predict the corresponding binding affinity using a scoring function. This method has great value in drug design. After decades of development, scoring functions nowadays typically can identify the true binding mode, but the prediction of binding affinity still remains a major problem. Here we present CScore, a data-driven scoring function using a modified Cerebellar Model Articulation Controller (CMAC) learning architecture, for accurate binding affinity prediction. The performance of CScore in terms of correlation between predicted and experimental binding affinities is benchmarked under different validation approaches. CScore achieves a prediction with R = 0.7668 and RMSE = 1.4540 when tested on an independent dataset. To the best of our knowledge, this result outperforms other scoring functions tested on the same dataset. The performance of CScore varies on different clusters under the leave-cluster-out validation approach, but still achieves competitive result. Lastly, the target-specified CScore achieves an even better result with R = 0.8237 and RMSE = 1.0872, trained on a much smaller but more relevant dataset for each target. The large dataset of protein-ligand complexes structural information and advances of machine learning techniques enable the data-driven approach in binding affinity prediction. CScore is capable of accurate binding affinity prediction. It is also shown that CScore will perform better if sufficient and relevant data is presented. As there is growth of publicly available structural data, further improvement of this scoring scheme can be expected.

Pyrone‐Based Inhibitors of Metalloproteinase Types 2 and 3 May Work as Conformation‐Selective Inhibitors

Matrix metalloproteinases are zinc-containing enzymes capable of degrading all components of the extracellular matrix. Owing to their role in human disease, matrix metalloproteinase have been the subject of extensive study. A bioinorganic approach was recently used to identify novel inhibitors based on a maltol zinc-binding group, but accompanying molecular-docking studies failed to explain why one of these inhibitors, AM-6, had approximately 2500-fold selectivity for MMP-3 over MMP-2. A number of studies have suggested that the matrix-metalloproteinase active site is highly flexible, leading some to speculate that differences in active-site flexibility may explain inhibitor selectivity. To extend the bioinorganic approach in a way that accounts for MMP-2 and MMP-3 dynamics, we here investigate the predicted binding modes and energies of AM-6 docked into multiple structures extracted from matrix-metalloproteinase molecular dynamics simulations. Our findings suggest that accounting for protein dynamics is essential for the accurate prediction of binding affinity and selectivity. Additionally, AM-6 and other similar inhibitors likely select for and stabilize only a subpopulation of all matrix-metalloproteinase conformations sampled by the apo protein. Consequently, when attempting to predict ligand affinity and selectivity using an ensemble of protein structures, it may be wise to disregard protein conformations that cannot accommodate the ligand.

Computational design of an endo-1,4- -xylanase ligand binding site

The field of computational protein design has experienced important recent success. However, the de novo computational design of high-affinity protein-ligand interfaces is still largely an open challenge. Using the Rosetta program, we attempted the in silico design of a high-affinity protein interface to a small peptide ligand. We chose the thermophilic endo-1,4-β-xylanase from Nonomuraea flexuosa as the protein scaffold on which to perform our designs. Over the course of the study, 12 proteins derived from this scaffold were produced and assayed for binding to the target ligand. Unfortunately, none of the designed proteins displayed evidence of high-affinity binding. Structural characterization of four designed proteins revealed that although the predicted structure of the protein model was highly accurate, this structural accuracy did not translate into accurate prediction of binding affinity. Crystallographic analyses indicate that the lack of binding affinity is possibly due to unaccounted for protein dynamics in the 'thumb' region of our design scaffold intrinsic to the family 11 β-xylanase fold. Further computational analysis revealed two specific, single amino acid substitutions responsible for an observed change in backbone conformation, and decreased dynamic stability of the catalytic cleft. These findings offer new insight into the dynamic and structural determinants of the β-xylanase proteins.

Formal Estimation of Errors in Computed Absolute Interaction Energies of Protein-ligand Complexes.

A largely unsolved problem in computational biochemistry is the accurate prediction of binding affinities of small ligands to protein receptors. We present a detailed analysis of the systematic and random errors present in computational methods through the use of error probability density functions, specifically for computed interaction energies between chemical fragments comprising a protein-ligand complex. An HIV-II protease crystal structure with a bound ligand (indinavir) was chosen as a model protein-ligand complex. The complex was decomposed into twenty-one (21) interacting fragment pairs, which were studied using a number of computational methods. The chemically accurate complete basis set coupled cluster theory (CCSD(T)/CBS) interaction energies were used as reference values to generate our error estimates. In our analysis we observed significant systematic and random errors in most methods, which was surprising especially for parameterized classical and semiempirical quantum mechanical calculations. After propagating these fragment-based error estimates over the entire protein-ligand complex, our total error estimates for many methods are large compared to the experimentally determined free energy of binding. Thus, we conclude that statistical error analysis is a necessary addition to any scoring function attempting to produce reliable binding affinity predictions.

Development of accurate binding affinity predictions of novel renin inhibitors through molecular docking studies

In this study, an attempt was made to explore a possible correlation between different docking scoring functions (Glide InducedFit docking score and GOLD's GoldScore and ChemScore) and binding energy values of a set of renin inhibitors, using linear regression model. The renin inhibitors under study are characterized by known bound to the receptor crystal structures possessing a great variety of pharmacophore groups and a wide range of IC 50 values. Linear regression models were derived to relate the docking scoring function and pIC 50 values of renin inhibitors under study. The developed derived models are seeking to be helpful for the rational design of new, more potent renin inhibitors.

Are Scoring Functions in Protein−Protein Docking Ready To Predict Interactomes? Clues from a Novel Binding Affinity Benchmark

The design of an ideal scoring function for protein-protein docking that would also predict the binding affinity of a complex is one of the challenges in structural proteomics. Such a scoring function would open the route to in silico, large-scale annotation and prediction of complete interactomes. Here we present a protein-protein binding affinity benchmark consisting of binding constants (K(d)'s) for 81 complexes. This benchmark was used to assess the performance of nine commonly used scoring algorithms along with a free-energy prediction algorithm in their ability to predicting binding affinities. Our results reveal a poor correlation between binding affinity and scores for all algorithms tested. However, the diversity and validity of the benchmark is highlighted when binding affinity data are categorized according to the methodology by which they were determined. By further classifying the complexes into low, medium and high affinity groups, significant correlations emerge, some of which are retained after dividing the data into more classes, showing the robustness of these correlations. Despite this, accurate prediction of binding affinity remains outside our reach due to the large associated standard deviations of the average score within each group. All the above-mentioned observations indicate that improvements of existing scoring functions or design of new consensus tools will be required for accurate prediction of the binding affinity of a given protein-protein complex. The benchmark developed in this work will serve as an indispensable source to reach this goal.

High-Throughput All-Atom Molecular Dynamics Simulations Using Distributed Computing

Although molecular dynamics simulation methods are useful in the modeling of macromolecular systems, they remain computationally expensive, with production work requiring costly high-performance computing (HPC) resources. We review recent innovations in accelerating molecular dynamics on graphics processing units (GPUs), and we describe GPUGRID, a volunteer computing project that uses the GPU resources of nondedicated desktop and workstation computers. In particular, we demonstrate the capability of simulating thousands of all-atom molecular trajectories generated at an average of 20 ns/day each (for systems of approximately 30 000-80 000 atoms). In conjunction with a potential of mean force (PMF) protocol for computing binding free energies, we demonstrate the use of GPUGRID in the computation of accurate binding affinities of the Src SH2 domain/pYEEI ligand complex by reconstructing the PMF over 373 umbrella sampling windows of 55 ns each (20.5 mus of total data). We obtain a standard free energy of binding of -8.7 +/- 0.4 kcal/mol within 0.7 kcal/mol from experimental results. This infrastructure will provide the basis for a robust system for high-throughput accurate binding affinity prediction.

Application of the PM6 semi-empirical method to modeling proteins enhances docking accuracy of AutoDock.

BackgroundMolecular docking methods are commonly used for predicting binding modes and energies of ligands to proteins. For accurate complex geometry and binding energy estimation, an appropriate method for calculating partial charges is essential. AutoDockTools software, the interface for preparing input files for one of the most widely used docking programs AutoDock 4, utilizes the Gasteiger partial charge calculation method for both protein and ligand charge calculation. However, it has already been shown that more accurate partial charge calculation - and as a consequence, more accurate docking- can be achieved by using quantum chemical methods. For docking calculations quantum chemical partial charge calculation as a routine was only used for ligands so far. The newly developed Mozyme function of MOPAC2009 allows fast partial charge calculation of proteins by quantum mechanical semi-empirical methods. Thus, in the current study, the effect of semi-empirical quantum-mechanical partial charge calculation on docking accuracy could be investigated.ResultsThe docking accuracy of AutoDock 4 using the original AutoDock scoring function was investigated on a set of 53 protein ligand complexes using Gasteiger and PM6 partial charge calculation methods. This has enabled us to compare the effect of the partial charge calculation method on docking accuracy utilizing AutoDock 4 software. Our results showed that the docking accuracy in regard to complex geometry (docking result defined as accurate when the RMSD of the first rank docking result complex is within 2 Å of the experimentally determined X-ray structure) significantly increased when partial charges of the ligands and proteins were calculated with the semi-empirical PM6 method.Out of the 53 complexes analyzed in the course of our study, the geometry of 42 complexes were accurately calculated using PM6 partial charges, while the use of Gasteiger charges resulted in only 28 accurate geometries. The binding affinity estimation was not influenced by the partial charge calculation method - for more accurate binding affinity prediction development of a new scoring function for AutoDock is needed.ConclusionOur results demonstrate that the accuracy of determination of complex geometry using AutoDock 4 for docking calculation greatly increases with the use of quantum chemical partial charge calculation on both the ligands and proteins.

Accurate approximation method for prediction of class I MHC affinities for peptides of length 8, 10 and 11 using prediction tools trained on 9mers

Several accurate prediction systems have been developed for prediction of class I major histocompatibility complex (MHC):peptide binding. Most of these are trained on binding affinity data of primarily 9mer peptides. Here, we show how prediction methods trained on 9mer data can be used for accurate binding affinity prediction of peptides of length 8, 10 and 11. The method gives the opportunity to predict peptides with a different length than nine for MHC alleles where no such peptides have been measured. As validation, the performance of this approach is compared to predictors trained on peptides of the peptide length in question. In this validation, the approximation method has an accuracy that is comparable to or better than methods trained on a peptide length identical to the predicted peptides. The algorithm has been implemented in the web-accessible servers NetMHC-3.0: http://www.cbs.dtu.dk/services/NetMHC-3.0, and NetMHCpan-1.1: http://www.cbs.dtu.dk/services/NetMHCpan-1.1

A new method to estimate ligand-receptor energetics.

In the discovery of new drugs, lead identification and optimization have assumed critical importance given the number of drug targets generated from genetic, genomics, and proteomic technologies. High-throughput experimental screening assays have been complemented recently by "virtual screening" approaches to identify and filter potential ligands when the characteristics of a target receptor structure of interest are known. Virtual screening mandates a reliable procedure for automatic ranking of structurally distinct ligands in compound library databases. Computing a rank score requires the accurate prediction of binding affinities between these ligands and the target. Many current scoring strategies require information about the target three-dimensional structure. In this study, a new method to estimate the free binding energy between a ligand and receptor is proposed. We extend a central idea previously reported (Bock, J. R., and Gough, D. A. (2001) Predicting protein-protein interactions from primary structure. Bioinformatics 17, 455-460; Bock, J. R., and Gough, D. A. (2002) Whole-proteome interaction mining. Bioinformatics, in press) that uses simple descriptors to represent biomolecules as input examples to train a support vector machine (Smola, A. J., and Schölkopf, B. (1998) A Tutorial on Support Vector Regression, Neuro-COLT Technical Report NC-TR-98-030, Royal Holloway College, University of London, UK) and the application of the trained system to previously unseen pairs, estimating their propensity for interaction. Here we seek to learn the function that maps features of a receptor-ligand pair onto their equilibrium free binding energy. These features do not comprise any direct information about the three-dimensional structures of ligand or target. In cross-validation experiments, it is demonstrated that objective measurements of prediction error rate and rank-ordering statistics are competitive with those of several other investigations, most of which depend on three-dimensional structural data. The size of the sample (n = 2,671) indicates that this approach is robust and may have widespread applicability beyond restricted families of receptor types. It is concluded that newly sequenced proteins, or those for which three-dimensional crystal structures are not easily obtained, can be rapidly analyzed for their binding potential against a library of ligands using this methodology.