Surflex-Dock: Docking benchmarks and real-world application

Benchmarks for molecular docking have historically focused on re-docking the cognate ligand of a well-determined protein-ligand complex to measure geometric pose prediction accuracy, and measurement of virtual screening performance has been focused on increasingly large and diverse sets of target protein structures, cognate ligands, and various types of decoy sets. Here, pose prediction is reported on the Astex Diverse set of 85 protein ligand complexes, and virtual screening performance is reported on the DUD set of 40 protein targets. In both cases, prepared structures of targets and ligands were provided by symposium organizers. The re-prepared data sets yielded results not significantly different than previous reports of Surflex-Dock on the two benchmarks. Minor changes to protein coordinates resulting from complex pre-optimization had large effects on observed performance, highlighting the limitations of cognate ligand re-docking for pose prediction assessment. Docking protocols developed for cross-docking, which address protein flexibility and produce discrete families of predicted poses, produced substantially better performance for pose prediction. Performance on virtual screening performance was shown to benefit by employing and combining multiple screening methods: docking, 2D molecular similarity, and 3D molecular similarity. In addition, use of multiple protein conformations significantly improved screening enrichment.     

A pose prediction approach based on ligand 3D shape similarity

Molecular docking predicts the best pose of a ligand in the target protein binding site by sampling and scoring numerous conformations and orientations of the ligand. Failures in pose prediction are often due to either insufficient sampling or scoring function errors. To improve the accuracy of pose prediction by tackling the sampling problem, we have developed a method of pose prediction using shape similarity. It first places a ligand conformation of the highest 3D shape similarity with known crystal structure ligands into protein binding site and then refines the pose by repacking the side-chains and performing energy minimization with a Monte Carlo algorithm. We have assessed our method utilizing CSARdock 2012 and 2014 benchmark exercise datasets consisting of co-crystal structures from eight proteins. Our results revealed that ligand 3D shape similarity could substitute conformational and orientational sampling if at least one suitable co-crystal structure is available. Our method identified poses within 2 Å RMSD as the top-ranking pose for 85.7 % of the test cases. The median RMSD for our pose prediction method was found to be 0.81 Å and was better than methods performing extensive conformational and orientational sampling within target protein binding sites. Furthermore, our method was better than similar methods utilizing ligand 3D shape similarity for pose prediction.     

An alternative method for the evaluation of docking performance: RSR vs RMSD

A new assessment criterion for docking poses is proposed in which experimental electron density is taken into account when evaluating the ability of docking programs to reproduce experimentally observed binding modes. Three docking programs (Gold, Glide, and Fred) were used to generate poses for a set of 88 protein-ligand complexes for which the crystal structure is known. The new criterion is based on the real space R-factor (RSR), which measures how well a group of atoms-in our case the ligand-fits the experimental electron density by comparing that density to the expected density, calculated from the model (i.e., the predicted ligand pose). The RSR-based measure is compared to the traditional criterion, the root-mean-square distance (RMSD) between the docking pose and the binding configuration in the crystallographic model. The results highlight several shortcomings of the RMSD criterion that do not affect the RSR-based measure. Examples illustrate that the RSR-derived approach allows a more meaningful a posteriori assessment of docking methods and results. Practical implications for docking evaluations and for methodological development work in this field are discussed.     

Machine learning optimization of cross docking accuracy

Performance of small molecule automated docking programs has conceptually been divided into docking -, scoring -, ranking - and screening power, which focuses on the crystal pose prediction, affinity prediction, ligand ranking and database screening capabilities of the docking program, respectively. Benchmarks show that different docking programs can excel in individual benchmarks which suggests that the scoring function employed by the programs can be optimized for a particular task. Here the scoring function of Smina is re-optimized towards enhancing the docking power using a supervised machine learning approach and a manually curated database of ligands and cross docking receptor pairs. The optimization method does not need associated binding data for the receptor-ligand examples used in the data set and works with small train sets. The re-optimization of the weights for the scoring function results in a similar docking performance with regard to docking power towards a cross docking test set. A ligand decoy based benchmark indicates a better discrimination between poses with high and low RMSD. The reported parameters for Smina are compatible with Autodock Vina and represent ready-to-use alternative parameters for researchers who aim at pose prediction rather than affinity prediction.     

Prediction of binding poses to FXR using multi-targeted docking combined with molecular dynamics and enhanced sampling

Advanced molecular docking methods often aim at capturing the flexibility of the protein upon binding to the ligand. In this study, we investigate whether instead a simple rigid docking method can be applied, if combined with multiple target structures to model the backbone flexibility and molecular dynamics simulations to model the sidechain and ligand flexibility. The methods are tested for the binding of 35 ligands to FXR as part of the first stage of the Drug Design Data Resource (D3R) Grand Challenge 2 blind challenge. The results show that the multiple-target docking protocol performs surprisingly well, with correct poses found for 21 of the ligands. MD simulations started on the docked structures are remarkably stable, but show almost no tendency of refining the structure closer to the experimentally found binding pose. Reconnaissance metadynamics enhances the exploration of new binding poses, but additional collective variables involving the protein are needed to exploit the full potential of the method.     

Assessing the role of polarization in docking

We describe a strategy for including ligand and protein polarization in docking that is based on the conversion of induced dipoles to induced charges. Induced charges have a distinct advantage in that they are readily implemented into a number of different computer programs, including many docking programs and hybrid QM/MM programs; induced charges are also more readily interpreted. In this study, the ligand was treated quantum mechanically to avoid parametrization issues and was polarized by the target protein, which was treated as a set of point charges. The induced dipole at a given target atom, due to polarization by the ligand and neighboring residues, was reformulated as induced charges at the given atom and its bonded neighbors, and these were allowed to repolarize the ligand in an iterative manner. The final set of polarized charges was evaluated in docking using AutoDock 4.0 on 12 protein-ligand systems against the default empirical Gasteiger charges, and against nonpolarized and partially polarized potential-derived charges. One advantage of AutoDock is that the best rmsd structure can be identified not only from the lowest energy pose but also from the largest cluster of poses. Inclusion of polarization does not always lead to the lowest energy pose having a lower rmsd, because docking is designed by necessity to be rapid rather than accurate. However, whenever an improvement in methodology, corresponding to a more thorough treatment of polarization, resulted in an increased cluster size, then there was also a corresponding decrease in the rmsd. The options for implementing polarization within a purely classical docking framework are discussed.     

Effects of protein conformation in docking: improved pose prediction through protein pocket adaptation

Computational methods for docking ligands have been shown to be remarkably dependent on precise protein conformation, where acceptable results in pose prediction have been generally possible only in the artificial case of re-docking a ligand into a protein binding site whose conformation was determined in the presence of the same ligand (the “cognate” docking problem). In such cases, on well curated protein/ligand complexes, accurate dockings can be returned as top-scoring over 75% of the time using tools such as Surflex-Dock. A critical application of docking in modeling for lead optimization requires accurate pose prediction for novel ligands, ranging from simple synthetic analogs to very different molecular scaffolds. Typical results for widely used programs in the “cross-docking case” (making use of a single fixed protein conformation) have rates closer to 20% success. By making use of protein conformations from multiple complexes, Surflex-Dock yields an average success rate of 61% across eight pharmaceutically relevant targets. Following docking, protein pocket adaptation and rescoring identifies single pose families that are correct an average of 67% of the time. Consideration of the best of two pose families (from alternate scoring regimes) yields a 75% mean success rate.     

Pose prediction accuracy in docking studies and enrichment of actives in the active site of GSK-3beta

We present molecular docking studies on the inhibitors of GSK-3beta kinase in the enzyme binding sites of the X-ray complexes (1H8F, 1PYX, 1O9U, 1Q4L, 1Q5K, and 1UV5) using the Schr?dinger docking tool Glide. Cognate and cross-docking studies using standard precision (SP) and extraprecision (XP) algorithms have been carried out. Cognate docking studies demonstrate that docked poses similar to X-ray poses (root-mean-square deviations of less than 2 A) are found within the top four ranks of the GlideScore and E-model scores. However, cross-docking studies typically produce poses that are significantly deviated from X-ray poses in all but a couple of cases, implying potential for induced fit effects in ligand binding. In this light, we have also carried out induced fit docking studies in the active sites of 1O9U, 1Q4L, and 1Q5K. Specifically, conformational changes have been effected in the active sites of these three protein structures to dock noncognate ligands. Thus, for example, the active site of 1O9U has been induced to fit the ligands of 1Q4L, 1Q5K, and 1UV5. These studies produce ligand docked poses which have significantly lower root-mean-square deviations relative to their X-ray crystallographic poses, when compared to the corresponding values from the cross-docking studies. Furthermore, we have used an ensemble of the induced fit models and X-ray structures to enhance the retrieval of active GSK-3beta inhibitors seeded in a decoy database, normally used in Glide validation studies. Thus, our studies provide valuable insights into computational strategies useful for the identification of potential GSK-3beta inhibitors.     

GalaxyDock BP2 score: a hybrid scoring function for accurate protein–ligand docking

Protein–ligand docking is a useful tool for providing atomic-level understanding of protein functions in nature and design principles for artificial ligands or proteins with desired properties. The ability to identify the true binding pose of a ligand to a target protein among numerous possible candidate poses is an essential requirement for successful protein–ligand docking. Many previously developed docking scoring functions were trained to reproduce experimental binding affinities and were also used for scoring binding poses. However, in this study, we developed a new docking scoring function, called GalaxyDock BP2 Score, by directly training the scoring power of binding poses. This function is a hybrid of physics-based, empirical, and knowledge-based score terms that are balanced to strengthen the advantages of each component. The performance of the new scoring function exhibits significant improvement over existing scoring functions in decoy pose discrimination tests. In addition, when the score is used with the GalaxyDock2 protein–ligand docking program, it outperformed other state-of-the-art docking programs in docking tests on the Astex diverse set, the Cross2009 benchmark set, and the Astex non-native set. GalaxyDock BP2 Score and GalaxyDock2 with this score are freely available at http://galaxy.seoklab.org/softwares/galaxydock.html.     

Docking rigid macrocycles using Convex-PL,AutoDock Vina,and RDKit in the D3R Grand Challenge 4

The D3R Grand Challenge 4 provided a brilliant opportunity to test macrocyclic docking protocols on a diverse high-quality experimental data. We participated in both pose and affinity prediction exercises. Overall, we aimed to use an automated structure-based docking pipeline built around a set of tools developed in our team. This exercise again demonstrated a crucial importance of the correct local ligand geometry for the overall success of docking. Starting from the second part of the pose prediction stage, we developed a stable pipeline for sampling macrocycle conformers. This resulted in the subangstrom average precision of our pose predictions. In the affinity prediction exercise we obtained average results. However, we could improve these when using docking poses submitted by the best predictors. Our docking tools including the Convex-PL scoring function are available at https://team.inria.fr/nano-d/software/.     

Are predefined decoy sets of ligand poses able to quantify scoring function accuracy?

Due to the large number of different docking programs and scoring functions available, researchers are faced with the problem of selecting the most suitable one when starting a structure-based drug discovery project. To guide the decision process, several studies comparing different docking and scoring approaches have been published. In the context of comparing scoring function performance, it is common practice to use a predefined, computer-generated set of ligand poses (decoys) and to reevaluate their score using the set of scoring functions to be compared. But are predefined decoy sets able to unambiguously evaluate and rank different scoring functions with respect to pose prediction performance? This question arose when the pose prediction performance of our piecewise linear potential derived scoring functions (Korb et al. in J Chem Inf Model 49:84–96, 2009) was assessed on a standard decoy set (Cheng et al. in J Chem Inf Model 49:1079–1093, 2009). While they showed excellent pose identification performance when they were used for rescoring of the predefined decoy conformations, a pronounced degradation in performance could be observed when they were directly applied in docking calculations using the same test set. This implies that on a discrete set of ligand poses only the rescoring performance can be evaluated. For comparing the pose prediction performance in a more rigorous manner, the search space of each scoring function has to be sampled extensively as done in the docking calculations performed here. We were able to identify relative strengths and weaknesses of three scoring functions (ChemPLP, GoldScore, and Astex Statistical Potential) by analyzing the performance for subsets of the complexes grouped by different properties of the active site. However, reasons for the overall poor performance of all three functions on this test set compared to other test sets of similar size could not be identified.     

Protein-ligand NOE matching: a high-throughput method for binding pose evaluation that does not require protein NMR resonance assignments

Given the three-dimensional (3D) structure of a protein, the binding pose of a ligand can be determined using distance restraints derived from assigned intra-ligand and protein-ligand nuclear Overhauser effects (NOEs). A primary limitation of this approach is the need for resonance assignments of the ligand-bound protein. We have developed an approach that utilizes data from 3D 13C-edited, 13C/15N-filtered HSQC-NOESY spectra for evaluating ligand binding poses without requiring protein NMR resonance assignments. Only the 1H NMR assignments of the bound ligand are essential. Trial ligand binding poses are generated by any suitable method (e.g., computational docking). For each trial binding pose, the 3D 13C-edited, 13C/15N-filtered HSQC-NOESY spectrum is predicted, and the predicted and observed patterns of protein-ligand NOEs are matched and scored using a fast, deterministic bipartite graph matching algorithm. The best scoring (lowest "cost") poses are identified. Our method can incorporate any explicit restraints or protein assignment data that are available, and many extensions of the basic procedure are feasible. Only a single sample is required, and the method can be applied to both slowly and rapidly exchanging ligands. The method was applied to three test cases: one complex involving muscle fatty acid-binding protein (mFABP) and two complexes involving the leukocyte function-associated antigen 1 (LFA-1) I-domain. Without using experimental protein NMR assignments, the method identified the known binding poses with good accuracy. The addition of experimental protein NMR assignments improves the results. Our "NOE matching" approach is expected to be widely applicable; i.e., it does not appear to depend on a fortuitous distribution of binding pocket residues.     

Development and validation of a modular, extensible docking program: DOCK 5

We report on the development and validation of a new version of DOCK. The algorithm has been rewritten in a modular format, which allows for easy implementation of new scoring functions, sampling methods and analysis tools. We validated the sampling algorithm with a test set of 114 protein-ligand complexes. Using an optimized parameter set, we are able to reproduce the crystal ligand pose to within 2 A of the crystal structure for 79% of the test cases using our rigid ligand docking algorithm with an average run time of 1 min per complex and for 72% of the test cases using our flexible ligand docking algorithm with an average run time of 5 min per complex. Finally, we perform an analysis of the docking failures in the test set and determine that the sampling algorithm is generally sufficient for the binding pose prediction problem for up to 7 rotatable bonds; i.e. 99% of the rigid ligand docking cases and 95% of the flexible ligand docking cases are sampled successfully. We point out that success rates could be improved through more advanced modeling of the receptor prior to docking and through improvement of the force field parameters, particularly for structures containing metal-based cofactors.     

NMR-assisted computational studies of peptidomimetic inhibitors bound in the hydrophobic pocket of HIV-1 glycoprotein 41

Due to the inherently flexible nature of a protein–protein interaction surface, it is difficult both to inhibit the association with a small molecule, and to predict how it might bind to the surface. In this study, we have examined small molecules that mediate the interaction between a WWI motif on the C-helix of HIV-1 glycoprotein-41 (gp41) and a deep hydrophobic pocket contained in the interior N-helical trimer. Association between these two components of gp41 leads to virus–cell and cell–cell fusion, which could be abrogated in the presence of an inhibitor that binds tightly in the pocket. We have studied a comprehensive combinatorial library of α-helical peptidomimetics, and found that compounds with strongly hydrophobic side chains had the highest affinity. Computational docking studies produced multiple possible binding modes due to the flexibility of both the binding site and the peptidomimetic compounds. We applied a transferred paramagnetic relaxation enhancement experiment to two selected members of the library, and showed that addition of a few experimental constraints enabled definitive identification of unique binding poses. Computational docking results were extremely sensitive to side chain conformations, and slight variations could preclude observation of the experimentally validated poses. Different receptor structures were required for docking simulations to sample the correct pose for the two compounds. The study demonstrated the sensitivity of predicted poses to receptor structure and indicated the importance of experimental verification when docking to a malleable protein–protein interaction surface.     

Surflex-Dock 2.1: Robust performance from ligand energetic modeling,ring flexibility,and knowledge-based search

The Surflex flexible molecular docking method has been generalized and extended in two primary areas related to the search component of docking. First, incorporation of a small-molecule force-field extends the search into Cartesian coordinates constrained by internal ligand energetics. Whereas previous versions searched only the alignment and acyclic torsional space of the ligand, the new approach supports dynamic ring flexibility and all-atom optimization of docked ligand poses. Second, knowledge of well established molecular interactions between ligand fragments and a target protein can be directly exploited to guide the search process. This offers advantages in some cases over the search strategy where ligand alignment is guided solely by a “protomol” (a pre-computed molecular representation of an idealized ligand). Results are presented on both docking accuracy and screening utility using multiple publicly available benchmark data sets that place Surflex’s performance in the context of other molecular docking methods. In terms of docking accuracy, Surflex-Dock 2.1 performs as well as the best available methods. In the area of screening utility, Surflex’s performance is extremely robust, and it is clearly superior to other methods within the set of cases for which comparative data are available, with roughly double the screening enrichment performance.     

Evaluation of library ranking efficacy in virtual screening

We present the results of a comprehensive study in which we explored how the docking procedure affects the performance of a virtual screening approach. We used four docking engines and applied 10 scoring functions to the top-ranked docking solutions of seeded databases against six target proteins. The scores of the experimental poses were placed within the total set to assess whether the scoring function required an accurate pose to provide the appropriate rank for the seeded compounds. This method allows a direct comparison of library ranking efficacy. Our results indicate that the LigandFit/Ligscore1 and LigandFit/GOLD docking/scoring combinations, and to a lesser degree FlexX/FlexX, Glide/Ligscore1, DOCK/PMF (Tripos implementation), LigandFit1/Ligscore2 and LigandFit/PMF (Tripos implementation) were able to retrieve the highest number of actives at a 10% fraction of the database when all targets were looked upon collectively. We also show that the scoring functions rank the observed binding modes higher than the inaccurate poses provided that the experimental poses are available. This finding stresses the discriminatory ability of the scoring algorithms, when better poses are available, and suggests that the number of false positives can be lowered with conformers closer to bioactive ones.     

Combining in silico and in cerebro approaches for virtual screening and pose prediction in SAMPL4

The SAMPL challenges provide an ideal opportunity for unbiased evaluation and comparison of different approaches used in computational drug design. During the fourth round of this SAMPL challenge, we participated in the virtual screening and binding pose prediction on inhibitors targeting the HIV-1 integrase enzyme. For virtual screening, we used well known and widely used in silico methods combined with personal in cerebro insights and experience. Regular docking only performed slightly better than random selection, but the performance was significantly improved upon incorporation of additional filters based on pharmacophore queries and electrostatic similarities. The best performance was achieved when logical selection was added. For the pose prediction, we utilized a similar consensus approach that amalgamated the results of the Glide-XP docking with structural knowledge and rescoring. The pose prediction results revealed that docking displayed reasonable performance in predicting the binding poses. However, prediction performance can be improved utilizing scientific experience and rescoring approaches. In both the virtual screening and pose prediction challenges, the top performance was achieved by our approaches. Here we describe the methods and strategies used in our approaches and discuss the rationale of their performances.     

D3R Grand Challenge 4: ligand similarity and MM-GBSA-based pose prediction and affinity ranking for BACE-1 inhibitors

The Drug Design Data Resource (D3R) Grand Challenges present an opportunity to assess, in the context of a blind predictive challenge, the accuracy and the limits of tools and methodologies designed to help guide pharmaceutical drug discovery projects. Here, we report the results of our participation in the D3R Grand Challenge 4 (GC4), which focused on predicting the binding poses and affinity ranking for compounds targeting the $$\beta$$-amyloid precursor protein (BACE-1). Our ligand similarity-based protocol using HYBRID (OpenEye Scientific Software) successfully identified poses close to the native binding mode for most of the ligands with less than 2 Å RMSD accuracy. Furthermore, we compared the performance of our HYBRID-based approach to that of AutoDock Vina and DOCK 6 and found that using a reference ligand to guide the docking process is a better strategy for pose prediction and helped HYBRID to perform better here. We also conducted end-point free energy estimates on molecules dynamics based ensembles of protein-ligand complexes using molecular mechanics combined with generalized Born surface area method (MM-GBSA). We found that the binding affinity ranking based on MM-GBSA scores have poor correlation with the experimental values. Finally, the main lessons from our participation in D3R GC4 are: (i) the generation of the macrocyclic conformers is a key step for successful pose prediction, (ii) the protonation states of the BACE-1 binding site should be treated carefully, (iii) the MM-GBSA method could not discriminate well between different predicted binding poses, and (iv) the MM-GBSA method does not perform well at predicting protein–ligand binding affinities here.     

VoteDock: Consensus docking method for prediction of protein–ligand interactions

Molecular recognition plays a fundamental role in all biological processes, and that is why great efforts have been made to understand and predict protein–ligand interactions. Finding a molecule that can potentially bind to a target protein is particularly essential in drug discovery and still remains an expensive and time‐consuming task. In silico, tools are frequently used to screen molecular libraries to identify new lead compounds, and if protein structure is known, various protein–ligand docking programs can be used. The aim of docking procedure is to predict correct poses of ligand in the binding site of the protein as well as to score them according to the strength of interaction in a reasonable time frame. The purpose of our studies was to present the novel consensus approach to predict both protein–ligand complex structure and its corresponding binding affinity. Our method used as the input the results from seven docking programs (Surflex, LigandFit, Glide, GOLD, FlexX, eHiTS, and AutoDock) that are widely used for docking of ligands. We evaluated it on the extensive benchmark dataset of 1300 protein–ligands pairs from refined PDBbind database for which the structural and affinity data was available. We compared independently its ability of proper scoring and posing to the previously proposed methods. In most cases, our method is able to dock properly approximately 20% of pairs more than docking methods on average, and over 10% of pairs more than the best single program. The RMSD value of the predicted complex conformation versus its native one is reduced by a factor of 0.5 Å. Finally, we were able to increase the Pearson correlation of the predicted binding affinity in comparison with the experimental value up to 0.5. © 2010 Wiley Periodicals, Inc. J Comput Chem 32: 568–581, 2011     

Multiple ligand simultaneous docking: Orchestrated dancing of ligands in binding sites of protein

Present docking methodologies simulate only one single ligand at a time during docking process. In reality, the molecular recognition process always involves multiple molecular species. Typical protein–ligand interactions are, for example, substrate and cofactor in catalytic cycle; metal ion coordination together with ligand(s); and ligand binding with water molecules. To simulate the real molecular binding processes, we propose a novel multiple ligand simultaneous docking (MLSD) strategy, which can deal with all the above processes, vastly improving docking sampling and binding free energy scoring. The work also compares two search strategies: Lamarckian genetic algorithm and particle swarm optimization, which have respective advantages depending on the specific systems. The methodology proves robust through systematic testing against several diverse model systems: E. coli purine nucleoside phosphorylase (PNP) complex with two substrates, SHP2NSH2 complex with two peptides and Bcl‐xL complex with ABT‐737 fragments. In all cases, the final correct docking poses and relative binding free energies were obtained. In PNP case, the simulations also capture the binding intermediates and reveal the binding dynamics during the recognition processes, which are consistent with the proposed enzymatic mechanism. In the other two cases, conventional single‐ligand docking fails due to energetic and dynamic coupling among ligands, whereas MLSD results in the correct binding modes. These three cases also represent potential applications in the areas of exploring enzymatic mechanism, interpreting noisy X‐ray crystallographic maps, and aiding fragment‐based drug design, respectively. © 2010 Wiley Periodicals, Inc. J Comput Chem, 2010