EADock: docking of small molecules into protein active sites with a multiobjective evolutionary optimization

In recent years, protein-ligand docking has become a powerful tool for drug development. Although several approaches suitable for high throughput screening are available, there is a need for methods able to identify binding modes with high accuracy. This accuracy is essential to reliably compute the binding free energy of the ligand. Such methods are needed when the binding mode of lead compounds is not determined experimentally but is needed for structure-based lead optimization. We present here a new docking software, called EADock, that aims at this goal. It uses an hybrid evolutionary algorithm with two fitness functions, in combination with a sophisticated management of the diversity. EADock is interfaced with the CHARMM package for energy calculations and coordinate handling. A validation was carried out on 37 crystallized protein-ligand complexes featuring 11 different proteins. The search space was defined as a sphere of 15 A around the center of mass of the ligand position in the crystal structure, and on the contrary to other benchmarks, our algorithm was fed with optimized ligand positions up to 10 A root mean square deviation (RMSD) from the crystal structure, excluding the latter. This validation illustrates the efficiency of our sampling strategy, as correct binding modes, defined by a RMSD to the crystal structure lower than 2 A, were identified and ranked first for 68% of the complexes. The success rate increases to 78% when considering the five best ranked clusters, and 92% when all clusters present in the last generation are taken into account. Most failures could be explained by the presence of crystal contacts in the experimental structure. Finally, the ability of EADock to accurately predict binding modes on a real application was illustrated by the successful docking of the RGD cyclic pentapeptide on the alphaVbeta3 integrin, starting far away from the binding pocket.     

Homology-Modelling Protein-Ligand Interactions: Allowing for Ligand-Induced Conformational Change

Current homology-modelling methods do not consider small molecules in their automated processes. Therefore, the development of a reliable tool for protein-ligand homology modelling is an important next step in generating plausible models for molecular interactions. Two automated protein-ligand homology-modelling strategies, requiring no expert knowledge from the user, are investigated here. Both employ the “induced fit” concept with flexibility in side chains and ligand. The most successful strategy superimposes the new ligand over the original ligand before homology modelling, allowing the new ligand to be taken into consideration during protein modelling (rather than after), facilitating conformational change in the local backbone if necessary. We show that this approach results in successful modelling of the ligand and key binding-site residues of angiotensin-converting enzyme 2 (ACE2) from its homologue ACE, which is not possible via conventional homology modelling or by homology modelling followed by docking. Several other difficult target complexes are also successfully modelled, reproducing native protein-ligand contacts with significantly different biological substrates and different binding-site conformations. These include the modelling of Cdk5 (cyclin-dependent kinase 5) from Cdk2, thymidine phosphorylase from a bacterial homologue, and dihydrofolate reductase from a recombinant variant with a markedly different inhibitor. In terms of average modelling quality across 82 targets, the ligand RMSD with respect to the experimental structure is 1.4 Å (and 2.0 Å for the protein binding site) for “easy” cases and 2.9 Å for the ligand (and 2.7 Å for the protein binding site) in “hard” cases. This demonstrates the importance of selecting an optimal template. Ligand-modelling accuracy is strongly dependent on target-template ligand structural similarity, rather than target-template sequence identity. However, protein-modelling accuracy is dependent on both. Our automated protein-ligand homology-modelling strategy generates a higher degree of accuracy than homology modelling followed by docking, generating an average ligand RMSD that is 1-2 Å better than docking with homology models.     

Discrete molecular dynamics distinguishes nativelike binding poses from decoys in difficult targets

Virtual screening is one of the major tools used in computer-aided drug discovery. In structure-based virtual screening, the scoring function is critical to identifying the correct docking pose and accurately predicting the binding affinities of compounds. However, the performance of existing scoring functions has been shown to be uneven for different targets, and some important drug targets have proven especially challenging. In these targets, scoring functions cannot accurately identify the native or near-native binding pose of the ligand from among decoy poses, which affects both the accuracy of the binding affinity prediction and the ability of virtual screening to identify true binders in chemical libraries. Here, we present an approach to discriminating native poses from decoys in difficult targets for which several scoring functions failed to correctly identify the native pose. Our approach employs Discrete Molecular Dynamics simulations to incorporate protein-ligand dynamics and the entropic effects of binding. We analyze a collection of poses generated by docking and find that the residence time of the ligand in the native and nativelike binding poses is distinctly longer than that in decoy poses. This finding suggests that molecular simulations offer a unique approach to distinguishing the native (or nativelike) binding pose from decoy poses that cannot be distinguished using scoring functions that evaluate static structures. The success of our method emphasizes the importance of protein-ligand dynamics in the accurate determination of the binding pose, an aspect that is not addressed in typical docking and scoring protocols.     

High quality binding modes in docking ligands to proteins

Multiple near-optimal conformations of protein-ligand complexes provide a better chance for accurate representation of biomolecular interactions, compared with a single structure. We present ISE-dock--a docking program which is based on the iterative stochastic elimination (ISE) algorithm. ISE eliminates values that consistently lead to the worst results, thus optimizing the search for docking poses. It constructs large sets of such poses with no additional computational cost compared with single poses. ISE-dock is validated using 81 protein-ligand complexes from the PDB and its performance was compared with those of Glide, GOLD, and AutoDock. ISE-dock has a better chance than the other three to find more than 60% top single poses under RMSD = 2.0 A and more than 80% under RMSD = 3.0 A from experimental. ISE alone produced at least one 3.0 A or better solutions among the top 20 poses in the entire test set. In 98% of the examined molecules, ISE produced solutions that are closer than 2.0 A from experimental. Paired t-tests (PTT) were used throughout to assess the significance of comparisons between the performances of the different programs. ISE-dock provides more than 100-fold docking solutions in a similar time frame as LGA in AutoDock. We demonstrate the usefulness of the large near optimal populations of ligand poses by showing a correlation between the docking results and experiments that support multiple binding modes in p38 MAP kinase (Pargellis et al., Nat Struct Biol 2002;9:268-272] and in Human Transthyretin (Hamilton, Benson, Cell Mol Life Sci 2001;58:1491-1521).     

ParDOCK: an all atom energy based Monte Carlo docking protocol for protein-ligand complexes

We report here an all-atom energy based Monte Carlo docking procedure tested on a dataset of 226 protein-ligand complexes. Average root mean square deviation (RMSD) from crystal conformation was observed to be approximately 0.53 A. The correlation coefficient (r(2)) for the predicted binding free energies calculated using the docked structures against experimental binding affinities was 0.72. The docking protocol is web-enabled as a free software at www.scfbio-iitd.res.in/dock.     

Selection and flexible optimization of binding modes from conformation ensembles

This paper describes a new semi-flexible docking approach named Fado (flexible alignment and docking), which incorporates flexibility by using an ensemble of precomputed ligand conformers. A primary ligand is defined as a linear combination over all input conformers. An optimization with regard to the linear coefficients makes the ligand flexible. Initially, a point matching problem utilizing the Merck Molecular Force Field (MMFF) is modeled in order to compute the correct orientation of the ligand with respect to the target. The problem is then solved through a local optimization approach (RPROP). This is done for 20 randomized ligand orientations, yielding 20 binding modes per complex. Evaluating these modes illustrates that our method is able to reproduce the binding modes of molecules within a few minutes of CPU time. A representative dataset of diverse protein-ligand complexes could be reproduced with 78% accuracy below 2A RMSD distance to the reference crystal structure. Fado is available upon request to the authors (see also http://www.zib.eu/Numerik/projects/docking/projectlong.en.html).     

PSI-DOCK: towards highly efficient and accurate flexible ligand docking

We have developed a new docking method, Pose-Sensitive Inclined (PSI)-DOCK, for flexible ligand docking. An improved SCORE function has been developed and used in PSI-DOCK for binding free energy evaluation. The improved SCORE function was able to reproduce the absolute binding free energies of a training set of 200 protein-ligand complexes with a correlation coefficient of 0.788 and a standard error of 8.13 kJ/mol. For ligand binding pose exploration, a unique searching strategy was designed in PSI-DOCK. In the first step, a tabu-enhanced genetic algorithm with a rapid shape-complementary scoring function is used to roughly explore and store potential binding poses of the ligand. Then, these predicted binding poses are optimized and compete against each other by using a genetic algorithm with the accurate SCORE function to determine the binding pose with the lowest docking energy. The PSI-DOCK 1.0 program is highly efficient in identifying the experimental binding pose. For a test dataset of 194 complexes, PSI-DOCK 1.0 achieved a 67% success rate (RMSD < 2.0 A) for only one run and a 74% success rate for 10 runs. PSI-DOCK can also predict the docking binding free energy with high accuracy. For a test set of 64 complexes, the correlation between the experimentally observed binding free energies and the docking binding free energies for 64 complexes is r = 0.777 with a standard deviation of 7.96 kJ/mol. Moreover, compared with other docking methods, PSI-DOCK 1.0 is extremely easy to use and requires minimum docking preparations. There is no requirement for the users to add hydrogen atoms to proteins because all protein hydrogen atoms and the flexibility of the terminal protein atoms are intrinsically taken into account in PSI-DOCK. There is also no requirement for the users to calculate partial atomic charges because PSI-DOCK does not calculate an electrostatic energy term. These features are not only convenient for the users but also help to avoid the influence of different preparation methods.     

An efficient computational method for calculating ligand binding affinities

Virtual compound screening using molecular docking is widely used in the discovery of new lead compounds for drug design. However, the docking scores are not sufficiently precise to represent the protein-ligand binding affinity. Here, we developed an efficient computational method for calculating protein-ligand binding affinity, which is based on molecular mechanics generalized Born/surface area (MM-GBSA) calculations and Jarzynski identity. Jarzynski identity is an exact relation between free energy differences and the work done through non-equilibrium process, and MM-GBSA is a semimacroscopic approach to calculate the potential energy. To calculate the work distribution when a ligand is pulled out of its binding site, multiple protein-ligand conformations are randomly generated as an alternative to performing an explicit single-molecule pulling simulation. We assessed the new method, multiple random conformation/MM-GBSA (MRC-MMGBSA), by evaluating ligand-binding affinities (scores) for four target proteins, and comparing these scores with experimental data. The calculated scores were qualitatively in good agreement with the experimental binding affinities, and the optimal docking structure could be determined by ranking the scores of the multiple docking poses obtained by the molecular docking process. Furthermore, the scores showed a strong linear response to experimental binding free energies, so that the free energy difference of the ligand binding (ΔΔG) could be calculated by linear scaling of the scores. The error of calculated ΔΔG was within ≈±1.5 kcal•mol⁻¹ of the experimental values. Particularly, in the case of flexible target proteins, the MRC-MMGBSA scores were more effective in ranking ligands than those generated by the MM-GBSA method using a single protein-ligand conformation. The results suggest that, owing to its lower computational costs and greater accuracy, the MRC-MMGBSA offers efficient means to rank the ligands, in the post-docking process, according to their binding affinities, and to compare these directly with the experimental values.     

Hit clustering can improve virtual fragment screening: CDK2 and PARP1 case studies

Virtual fragment screening could be a promising alternative to existing experimental screening techniques. However, reliable methods of in silico fragment screening are yet to be established and validated. In order to develop such an approach we first checked how successful the existing molecular docking methods can be in predicting fragment binding affinities and poses. Using our Lead Finder docking software the RMSD of the binding energy prediction was observed to be 1.35 kcal/mol(-1) on a set of 26 experimentally characterized fragment inhibitors, and the RMSD of the predicted binding pose from the experimental one was <1.5 ?. Then, we explored docking of 68 fragments obtained from 39 drug molecules for which co-crystal structures were available from the PDB. It appeared that fragments that participate in oriented non-covalent interactions, such as hydrogen bonds and metal coordination, could be correctly docked in 70-80% of cases suggesting the potential success of rediscovering of corresponding drugs by in silico fragment approach. Based on these findings we've developed a virtual fragment screening technique which involved structural filtration of protein-ligand complexes for specific interactions and subsequent clustering in order to minimize the number of preferable starting fragment candidates. Application of this method led to 2 millimolar-scale fragment PARP1 inhibitors with a new scaffold.     

Energetic contributions of amino acid residues and its cross-talk to delineate ligand-binding mechanism

Receptor-based QSAR approaches can enumerate the energetic contributions of amino acid residues toward ligand binding only when experimental binding affinity is associated. The structural data of protein-ligand complexes are witnessing a tremendous growth in the Protein Data Bank deposited with a few entries on binding affinity. We present here a new approach to compute the E nergetic CONT ributions of A mino acid residues and its possible C ross-T alk (ECONTACT) to study ligand binding using per-residue energy decomposition, molecular dynamics simulations and rescoring method without the need for experimental binding affinity. This approach recognizes potential cross-talks among amino acid residues imparting a nonadditive effect to the binding affinity with evidence of correlative motions in the dynamics simulations. The protein-ligand interaction energies deduced from multiple structures are decomposed into per-residue energy terms, which are employed as variables to principal component analysis and generated cross-terms. Out of 16 cross-talks derived from eight datasets of protein-ligand systems, the ECONTACT approach is able to associate 10 potential cross-talks with site-directed mutagenesis, free energy, and dynamics simulations data strongly. We modeled these key determinants of ligand binding using joint probability density function (jPDF) to identify cross-talks in protein structures. The top two cross-talks identified by ECONTACT approach corroborated with the experimental findings. Furthermore, virtual screening exercise using ECONTACT models better discriminated known inhibitors from decoy molecules. This approach proposes the jPDF metric to estimate the probability of observing cross-talks in any protein-ligand complex. The source code and related resources to perform ECONTACT modeling is available freely at https://www.gujaratuniversity.ac.in/econtact /.     

Knowledge-based scoring function to predict protein-ligand interactions

The development and validation of a new knowledge-based scoring function (DrugScore) to describe the binding geometry of ligands in proteins is presented. It discriminates efficiently between well-docked ligand binding modes (root-mean-square deviation <2.0 A with respect to a crystallographically determined reference complex) and those largely deviating from the native structure, e.g. generated by computer docking programs. Structural information is extracted from crystallographically determined protein-ligand complexes using ReLiBase and converted into distance-dependent pair-preferences and solvent-accessible surface (SAS) dependent singlet preferences for protein and ligand atoms. Definition of an appropriate reference state and accounting for inaccuracies inherently present in experimental data is required to achieve good predictive power. The sum of the pair preferences and the singlet preferences is calculated based on the 3D structure of protein-ligand binding modes generated by docking tools. For two test sets of 91 and 68 protein-ligand complexes, taken from the Protein Data Bank (PDB), the calculated score recognizes poses generated by FlexX deviating <2 A from the crystal structure on rank 1 in three quarters of all possible cases. Compared to FlexX, this is a substantial improvement. For ligand geometries generated by DOCK, DrugScore is superior to the "chemical scoring" implemented into this tool, while comparable results are obtained using the "energy scoring" in DOCK. None of the presently known scoring functions achieves comparable power to extract binding modes in agreement with experiment. It is fast to compute, regards implicitly solvation and entropy contributions and produces correctly the geometry of directional interactions. Small deviations in the 3D structure are tolerated and, since only contacts to non-hydrogen atoms are regarded, it is independent from assumptions of protonation states.     

The particle concept: placing discrete water molecules during protein-ligand docking predictions.

Water is known to play a significant role in the formation of protein-ligand complexes. In this paper, we focus on the influence of water molecules on the structure of protein-ligand complexes. We present an algorithmic approach, called the particle concept, for integrating the placement of single water molecules in the docking algorithm of FLEXX. FLEXX is an incremental construction approach to ligand docking consisting of three phases: the selection of base fragments, the placement of the base fragments, and the incremental reconstruction of the ligand inside the active site of a protein. The goal of the extension is to find water molecules at favorable places in the protein-ligand interface which may guide the placement of the ligand. In a preprocessing phase, favorable positions of water molecules inside the active site are calculated and stored in a list of possible water positions. During the incremental construction phase, water molecules are placed at the precomputed positions if they can form additional hydrogen bonds to the ligand. Steric constraints resulting from the water molecules as well as the geometry of the hydrogen bonds are used to optimize the ligand orientation in the active site during the reconstruction process. We have tested the particle concept on a series of 200 protein-ligand complexes. Although the average improvement of the prediction results is minor, we were able to predict water molecules between the protein and the ligand correctly in several cases. For instance in the case of HIV-1 protease, where a single water molecule between the protein and the ligand is known to be of importance in complex formation, significant improvements can be achieved.     

The utility of geometrical and chemical restraint information extracted from predicted ligand-binding sites in protein structure refinement

Exhaustive exploration of molecular interactions at the level of complete proteomes requires efficient and reliable computational approaches to protein function inference. Ligand docking and ranking techniques show considerable promise in their ability to quantify the interactions between proteins and small molecules. Despite the advances in the development of docking approaches and scoring functions, the genome-wide application of many ligand docking/screening algorithms is limited by the quality of the binding sites in theoretical receptor models constructed by protein structure prediction. In this study, we describe a new template-based method for the local refinement of ligand-binding regions in protein models using remotely related templates identified by threading. We designed a Support Vector Regression (SVR) model that selects correct binding site geometries in a large ensemble of multiple receptor conformations. The SVR model employs several scoring functions that impose geometrical restraints on the Cα positions, account for the specific chemical environment within a binding site and optimize the interactions with putative ligands. The SVR score is well correlated with the RMSD from the native structure; in 47% (70%) of the cases, the Pearson’s correlation coefficient is >0.5 (>0.3). When applied to weakly homologous models, the average heavy atom, local RMSD from the native structure of the top-ranked (best of top five) binding site geometries is 3.1 Å (2.9 Å) for roughly half of the targets; this represents a 0.1 (0.3) Å average improvement over the original predicted structure. Focusing on the subset of strongly conserved residues, the average heavy atom RMSD is 2.6 Å (2.3 Å). Furthermore, we estimate the upper bound of template-based binding site refinement using only weakly related proteins to be ～2.6 Å RMSD. This value also corresponds to the plasticity of the ligand-binding regions in distant homologues. The Binding Site Refinement (BSR) approach is available to the scientific community as a web server that can be accessed at http://cssb.biology.gatech.edu/bsr/.     

Accounting for global protein deformability during protein-protein and protein-ligand docking

Computational docking methods are valuable tools aimed to simplify the costly process of drug development and improvement. Most current approaches assume a rigid receptor structure to allow virtual screening of large numbers of possible ligands and putative binding sites on a receptor molecule. However, inclusion of receptor flexibility can be of critical importance since binding of a ligand can lead to changes in the receptor protein conformation that are sterically necessary to accommodate a ligand. Recent approaches to efficiently account for receptor flexibility during docking simulations are reviewed. In particular, accounting efficiently for global conformational changes of the protein backbone during docking is a still challenging unsolved problem. An approximate method has recently been suggested that is based on relaxing the receptor conformation during docking in pre-calculated soft collective degrees of freedom (M. Zacharias, Rapid protein-ligand docking using soft modes from molecular dynamics simulations to account for protein deformability: binding of FK506 to FKBP, Proteins: Struct., Funct., Genet. 54 (2004) 759-767). Test applications on protein-protein docking and on docking the inhibitor staurosporine to the apo-form of cAMP-dependent protein kinase A catalytic domain indicate significant improvement of docking results compared to rigid docking at a very modest computational demand. Accounting for receptor conformational changes in pre-calculated global degrees of freedom might offer a promising route to improve systematic docking screening simulations.     

Prediction of ligand-binding sites of proteins by molecular docking calculation for a random ligand library

A new approach to predicting the ligand-binding sites of proteins was developed, using protein-ligand docking computation. In this method, many compounds in a random library are docked onto the whole protein surface. We assumed that the true ligand-binding site would exhibit stronger affinity to the compounds in the random library than the other sites, even if the random library did not include the ligand corresponding to the true binding site. We also assumed that the affinity of the true ligand-binding site would be correlated to the docking scores of the compounds in the random library, if the ligand-binding site was correctly predicted. We call this method the molecular-docking binding-site finding (MolSite) method. The MolSite method was applied to 89 known protein-ligand complex structures extracted from the Protein Data Bank, and it predicted the correct binding sites with about 80-99% accuracy, when only the single top-ranked site was adopted. In addition, the average docking score was weakly correlated to the experimental protein-ligand binding free energy, with a correlation coefficient of 0.44.     

Implicit flexibility in protein docking: cross-docking and local refinement

In previous CAPRI rounds (3-5) we showed that using MD-generated ensembles, as inputs for a rigid-body docking algorithm, increased our success rate, especially for targets exhibiting substantial amounts of induced fit. In recent rounds (6-11), our cross-docking was followed by a short MD-based local refinement for the subset of solutions with the lowest interaction energies after minimization. The above approach showed promising results for target 20, where we were able to recover 30% of native contacts for one of our submitted models. Further tests, performed a posteriori, revealed that cross-docking approach produces more near-native (NN) solutions but only for targets with large conformational changes upon binding. However, at the time of the blind docking experiment, these improved solutions were not chosen for the subsequent refinement, as their interaction energies after minimization ranked poorly compared with other solutions. This indicates deficiencies in the present scoring schemes that are based on interaction energies of minimized structures. Refinement MD simulations substantially increase the fraction of native contacts for NN docked solutions, but generally worsen interface and ligand RMSD. Further analysis shows that although MD simulations are able to improve sidechain packing across the interface, which results in an increased fraction of native contacts, they are not capable of improving interface and ligand backbone RMSD for NN structures beyond 1.5 and 3.5 A, respectively, even if explicit solvent is used.     

Protein-ligand docking: current status and future challenges

Understanding the ruling principles whereby protein receptors recognize, interact, and associate with molecular substrates and inhibitors is of paramount importance in drug discovery efforts. Protein-ligand docking aims to predict and rank the structure(s) arising from the association between a given ligand and a target protein of known 3D structure. Despite the breathtaking advances in the field over the last decades and the widespread application of docking methods, several downsides still exist. In particular, protein flexibility-a critical aspect for a thorough understanding of the principles that guide ligand binding in proteins-is a major hurdle in current protein-ligand docking efforts that needs to be more efficiently accounted for. In this review the key concepts of protein-ligand docking methods are outlined, with major emphasis being given to the general strengths and weaknesses that presently characterize this methodology. Despite the size of the field, the principal types of search algorithms and scoring functions are reviewed and the most popular docking tools are briefly depicted. Recent advances that aim to address some of the traditional limitations associated with molecular docking are also described. A selection of hand-picked examples is used to illustrate these features.     

Molecular docking of balanol to dynamics snapshots of protein kinase A

Even if the structure of a receptor has been determined experimentally, it may not be a conformation to which a ligand would bind when induced fit effects are significant. Molecular docking using such a receptor structure may thus fail to recognize a ligand to which the receptor can bind with reasonable affinity. Here, we examine one way to alleviate this problem by using an ensemble of receptor conformations generated from a molecular dynamics simulation for molecular docking. Two molecular dynamics simulations were conducted to generate snapshots for protein kinase A: one with the ligand bound, the other without. The ligand, balanol, was then docked to conformations of the receptors presented by these trajectories. The Lamarckian genetic algorithm in Autodock [Goodsell et al. J Mol Recognit 1996;9(1):1-5; Morris et al. J Comput Chem 1998;19(14):1639-1662] was used in the docking. Three ligand models were used: rigid, flexible, and flexible with torsional potentials. When the snapshots were taken from the molecular dynamics simulation of the protein-ligand complex, the correct docking structure could be recovered easily by the docking algorithm in all cases. This was an easier case for challenging the docking algorithm because, by using the structure of the protein in a protein-ligand complex, one essentially assumed that the protein already had a pocket to which the ligand can fit well. However, when the snapshots were taken from the ligand-free protein simulation, which is more useful for a practical application when the structure of the protein-ligand complex is not known, several clusters of structures were found. Of the 10 docking runs for each snapshot, at least one structure was close to the correctly docked structure when the flexible-ligand models were used. We found that a useful way to identify the correctly docked structure was to locate the structure that appeared most frequently as the lowest energy structure in the docking experiments to different snapshots.