Novel statistical-thermodynamic methods to predict protein-ligand binding positions using probability distribution functions

We present two novel methods to predict native protein-ligand binding positions. Both methods identify the native binding position as the most probable position corresponding to a maximum of a probability distribution function (PDF) of possible binding positions in a protein active site. Possible binding positions are the origins of clusters composed, on the basis of root-mean square deviations (RMSD), from the multiple ligand positions determined by a docking algorithm. The difference between the methods lies in the ways the PDF is derived. To validate the suggested methods, we compare the averaged RMSD of the predicted ligand docked positions relative to the experimentally determined positions for a set of 135 PDB protein-ligand complexes. We demonstrate that the suggested methods improve docking accuracy by as much as 21-24% in comparison with a method that simply identifies the binding position as the energy top-scored ligand position.     

A new test set for validating predictions of protein-ligand interaction

We present a large test set of protein-ligand complexes for the purpose of validating algorithms that rely on the prediction of protein-ligand interactions. The set consists of 305 complexes with protonation states assigned by manual inspection. The following checks have been carried out to identify unsuitable entries in this set: (1) assessing the involvement of crystallographically related protein units in ligand binding; (2) identification of bad clashes between protein side chains and ligand; and (3) assessment of structural errors, and/or inconsistency of ligand placement with crystal structure electron density. In addition, the set has been pruned to assure diversity in terms of protein-ligand structures, and subsets are supplied for different protein-structure resolution ranges. A classification of the set by protein type is available. As an illustration, validation results are shown for GOLD and SuperStar. GOLD is a program that performs flexible protein-ligand docking, and SuperStar is used for the prediction of favorable interaction sites in proteins. The new CCDC/Astex test set is freely available to the scientific community (http://www.ccdc.cam.ac.uk).     

Evaluation of the FLEXX incremental construction algorithm for protein-ligand docking

We report on a test of FLEXX, a fully automatic docking tool for flexible ligands, on a highly diverse data set of 200 protein-ligand complexes from the Protein Data Bank. In total 46.5% of the complexes of the data set can be reproduced by a FLEXX docking solution at rank 1 with an rms deviation (RMSD) from the observed structure of less than 2 A. This rate rises to 70% if one looks at the entire generated solution set. FLEXX produces reliable results for ligands with up to 15 components which can be docked in 80% of the cases with acceptable accuracy. Ligands with more than 15 components tend to generate wrong solutions more often. The average runtime of FLEXX on this test set is 93 seconds per complex on a SUN Ultra-30 workstation. In addition, we report on "cross-docking" experiments, in which several receptor structures of complexes with identical proteins have been used for docking all cocrystallized ligands of these complexes. In most cases, these experiments show that FLEXX can acceptably dock a ligand into a foreign receptor structure. Finally we report on screening runs of ligands out of a library with 556 entries against ten different proteins. In eight cases FLEXX is able to find the original inhibitor within the top 7% of the total library.     

Efficacy and selectivity in flexible database docking

Flexible database docking with DOCK 4.0 has been evaluated for its ability to retrieve biologically active molecules from a database of approximately 1,000 compounds with known activities against thrombin and the progesterone receptor. The retrieval of known actives and chemically similar but inactive molecules was monitored as a function of conformational and orientational sampling. The largest enrichment of actives among the 10% highest ranking molecules is obtained when only five conformations are used to seed the next round of ligand reconstruction and limited sampling is applied to place the base fragment in the binding site. The performance of energy and chemical scoring, as implemented in DOCK 4.0, was found to depend on the protein used for docking. For the progesterone receptor, energy scoring yields the largest enrichments (64%) in terms of actives retrieved among the 10% top scoring molecules, while chemical scoring performs best for thrombin (94%). With the exception of the application of energy scoring to the progesterone receptor, both energy-based scoring schemes applied in this study do not discriminate well between true actives and chemically similar but inactive compounds. In conclusion, flexible docking is able to effectively prioritize high-throughput screening databases, using less conformational sampling than normally required for appropriate reconstruction of protein-ligand complexes. The more subtle discrimination between chemically similar classes of active and inactive compounds remains, however, problematic.     

Fragment-Based flexible ligand docking by evolutionary optimization.

A new computational approach for the efficient docking of flexible ligands in a rigid protein is presented. It exploits the binding modes of functional groups determined by an exhaustive search with solvation. The search in ligand conformational space is performed by a genetic algorithm whose scoring function approximates steric effects and intermolecular hydrogen bonds. Ligand conformations generated by the genetic algorithm are docked in the protein binding site by optimizing the fit of their fragments to optimal positions of chemically related functional groups. We show that the use of optimal binding modes of molecular fragments allows to dock known inhibitors with about ten rotatable bonds in the active site of the uncomplexed and complexed conformations of thrombin and HIV-1 protease.     

Docking of hydrophobic ligands with interaction-based matching algorithms

MOTIVATION: Matching of chemical interacting groups is a common concept for docking and fragment placement algorithms in computer-aided drug design. These algorithms have been proven to be reliable and fast if at least a certain number of hydrogen bonds or salt bridges occur. However, the algorithms typically run into problems if hydrophobic fragments or ligands should be placed. In order to dock hydrophobic fragments without significant loss of computational efficiency, we have extended the interaction model and placement algorithms in our docking tool FlexX. The concept of multi-level interactions is introduced into the algorithms for automatic selection and placement of base fragments. RESULTS: With the multi-level interaction model and the corresponding algorithmic extensions, we were able to improve the overall performance of FlexX significantly. We tested the approach with a set of 200 protein-ligand complexes taken from the Brookhaven Protein Data Bank (PDB). The number of test cases which can be docked within 1.5 A RMSD from the crystal structure can be increased from 58 to 64%. The performance gain is paid for by an increase in computation time from 73 to 91 s on average per protein-ligand complex. AVAILABILITY: The FlexX molecular docking software is available for UNIX platforms IRIX, Solaris and Linux. See http://cartan.gmd.de/FlexX for additional information.     

Towards Ligand Docking Including Explicit Interface Water Molecules

Small molecule docking predicts the interaction of a small molecule ligand with a protein at atomic-detail accuracy including position and conformation the ligand but also conformational changes of the protein upon ligand binding. While successful in the majority of cases, docking algorithms including RosettaLigand fail in some cases to predict the correct protein/ligand complex structure. In this study we show that simultaneous docking of explicit interface water molecules greatly improves Rosetta’s ability to distinguish correct from incorrect ligand poses. This result holds true for both protein-centric water docking wherein waters are located relative to the protein binding site and ligand-centric water docking wherein waters move with the ligand during docking. Protein-centric docking is used to model 99 HIV-1 protease/protease inhibitor structures. We find protease inhibitor placement improving at a ratio of 9∶1 when one critical interface water molecule is included in the docking simulation. Ligand-centric docking is applied to 341 structures from the CSAR benchmark of diverse protein/ligand complexes [1]. Across this diverse dataset we see up to 56% recovery of failed docking studies, when waters are included in the docking simulation.     

Analysis of structural water and CH···π interactions in HIV-1 protease and PTP1B complexes using a hydrogen bond prediction tool,HBPredicT

A hydrogen bond prediction tool HBPredicT is developed for detecting structural water molecules and CH···π interactions in PDB files of protein-ligand complexes. The program adds the missing hydrogen atoms to the protein, ligands, and oxygen atoms of water molecules and subsequently all the hydrogen bonds in the complex are located using specific geometrical criteria. Hydrogen bonds are classified into various types based on (i) donor and acceptor atoms, and interactions such as (ii) protein-protein, (iii) protein-ligand, (iv) protein-water, (v) ligand-water, (vi) water-water, and (vii) protein-water-ligand. Using the information in category (vii), the water molecules which form hydrogen bonds with the ligand and the protein simultaneously–the structural water–is identified and retrieved along with the associated ligand and protein residues. For CH···π interactions, the relevant portions of the corresponding structures are also extracted in the output. The application potential of this program is tested using 19 HIV-1 protease and 11 PTP1B inhibitor complexes. All the systems showed presence of structural water molecules and in several cases, the CH···π interaction between ligand and protein are detected. A rare occurrence of CH···π interactions emanating from both faces of a phenyl ring of the inhibitor is identified in HIV-1 protease 1D4L.     

A pseudo-particle approach for studying protein-ligand models truncated to their active sites

A molecular dynamics method has been developed to describe the structural and dynamic properties of protein-ligand complexes that are truncated to their active sites. The active site is comprised of the ligand and discontinuous, positionally unrestrained peptide chains. This truncated active-site complex is surrounded by big unspecific pseudo-particles representing the complete protein and the solvent. Thus, knowledge of the folding of the outer parts of the protein is not required, and the method can be applied to protein models, derived from homology modeling. The method has been tested using ligand complexes of adenylate kinase, retinol binding protein, HIV-1 protease, and human leucocyte antigen. Comparisons with their crystal structures and with results from time-demanding simulations of the whole complexes in explicit water solvent show that the ligand binding properties are conserved. Most of the hydrogen bonds between the ligand and the active-site residues are reproduced and, furthermore, the simulation time is reduced. © 1996 John Wiley & Sons, Inc.     

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.     

SuperStar: a knowledge-based approach for identifying interaction sites in proteins.

An empirical method for identifying interaction sites in proteins is described and validated. The method is based entirely on experimental information about non-bonded interactions occurring in small-molecule crystal structures. These data are used in the form of scatterplots that show the experimentally observed distribution of one functional group (the "contact group" or "probe") around another. A template molecule (e.g. a protein binding site) is broken down into structure fragments and the scatterplots, showing the distribution of a chosen probe around these structure fragments, are superimposed on the corresponding parts of the template. The scatterplots are then translated into a three-dimensional map that shows the propensity of the probe at different positions around the template molecule. The method is illustrated for l -arabinose-binding protein, complexed with l -arabinose and with d -fucose, and for dihydrofolate reductase complexed with methotrexate. The method is validated on 122 X-ray structures of protein-ligand complexes. For all the binding sites of these proteins, propensity maps are generated for four different probes: a charged NH+3nitrogen, a carbonyl oxygen, a hydroxyl oxygen and a methyl carbon atom. Next, the maps are compared with the experimentally observed positions of ligand atoms of these types. For 74% of these ligand atoms (84% of the solvent-inaccessible ones) the calculated propensity of the matching probe at the experimental positions is higher than expected by chance. For 68% of the atoms (82% of the solvent-inaccessible ones) the propensity of the matching probe is higher than that of the other three probes. These results indicate that the approach generally gives good predictions for protein-ligand interactions. The potential applications of the propensity maps range from an aid in manual docking and structure-based drug design to their use in pharmacophore development.     

Computational analysis of BACE1-ligand complex crystal structures and linear discriminant analysis for identification of BACE1 inhibitors with anti P-glycoprotein binding property

More than 100 years of research on Alzheimer’s disease didn’t yield a potential cure for this dreadful disease. Poor Blood Brain Barrier (BBB) permeability and P-glycoprotein binding of BACE1 inhibitors are the major causes for the failure of these molecules during clinical trials. The design of BACE1 inhibitors with a balance of sufficient affinity to the binding site and little or no interaction with P-glycoproteins is indispensable. Identification and understanding of protein–ligand interactions are essential for ligand optimization process. Structure-based drug design (SBDD) efforts led to a steady accumulation of BACE1-ligand crystal complexes in the PDB. This study focuses on analyses of 153 BACE1-ligand complexes for the direct contacts (hydrogen bonds and weak interactions) observed between protein and ligand and indirect contacts (water-mediated hydrogen bonds), observed in BACE1-ligand complex crystal structures. Intraligand hydrogen bonds were analyzed, with focus on ligand P-glycoprotein efflux. The interactions are dissected specific to subsites in the active site and discussed. The observed protein-ligand and intraligand interactions were used to develop the linear discriminant model for the identification of BACE1 inhibitors with less or no P-glycoprotein binding property. Excellent statistical results and model’s ability to correctly predict a new data-set with an accuracy of 92% is achieved. The results are retrospectively analyzed to give input for the design of potential BACE1 inhibitors.     

Estimating binding affinities by docking/scoring methods using variable protonation states

To investigate the effects of multiple protonation states on protein-ligand recognition, we generated alternative protonation states for selected titratable groups of ligands and receptors. The selection of states was based on the predicted pK(a) of the unbound receptor and ligand and the proximity of titratable groups of the receptor to the binding site. Various ligand tautomer states were also considered. An independent docking calculation was run for each state. Several protocols were examined: using an ensemble of all generated states of ligand and receptor, using only the most probable state of the unbound ligand/receptor, and using only the state giving the most favorable docking score. The accuracies of these approaches were compared, using a set of 176 protein-ligand complexes (15 receptors) for which crystal structures and measured binding affinities are available. The best agreement with experiment was obtained when ligand poses from experimental crystal structures were used. For 9 of 15 receptors, using an ensemble of all generated protonation states of the ligand and receptor gave the best correlation between calculated and measured affinities.     

SuperStar: comparison of CSD and PDB-based interaction fields as a basis for the prediction of protein-ligand interactions.

SuperStar is an empirical method for identifying interaction sites in proteins, based entirely on the experimental information about non-bonded interactions, present in the IsoStar database. The interaction information in IsoStar is contained in scatterplots, which show the distribution of a chosen probe around structure fragments. SuperStar breaks a template molecule (e.g. a protein binding site) into structural fragments which correspond to those in the scatterplots. The scatterplots are then superimposed on the corresponding parts of the template and converted into a composite propensity map.The original version of SuperStar was based entirely on scatterplots from the CSD. Here, scatterplots based on protein-ligand interactions are implemented in SuperStar, and validated on a test set of 122 X-ray structures of protein-ligand complexes. In this validation, propensity maps are compared with the experimentally observed positions of ligand atoms of comparable types. Although non-bonded interaction geometries in small molecule structures are similar to those found in protein-ligand complexes, their relative frequencies of occurrence are different. Polar interactions are more common in the first class of structures, while interactions between hydrophobic groups are more common in protein crystals. In general, PDB and CSD-based SuperStar maps appear equally successful in the prediction of protein-ligand interactions. PDB-based maps are more suitable to identify hydrophobic pockets, and inherently take into account the experimental uncertainties of protein atomic positions. If the protonation state of a histidine, aspartate or glutamate protein side-chain is known, specific CSD-based maps for that protonation state are preferred over PDB-based maps which represent an ensemble of protonation states.     

Thermodynamics of water mediating protein-ligand interactions in cytochrome P450cam: a molecular dynamics study.

Ordered water molecules are observed by crystallography and nuclear magnetic resonance to mediate protein-ligand interactions. Here, we examine the energetics of hydrating cavities formed at protein-ligand interfaces using molecular dynamics simulations. The free energies of hydrating two cavities in the active site of two liganded complexes of cytochrome P450cam were calculated by multiconfigurational thermodynamic integration. The complex of cytochrome P450cam with 2-phenyl-imidazole contains a crystallographically well defined water molecule mediating hydrogen bonds between the protein and the inhibitor. We calculate that this water molecule is stabilized by a binding free energy of -11.6 +/- kJ/mol. The complex of cytochrome P450cam with its natural substrate, camphor, contains a cavity that is empty in the crystal structure although a water molecule in it could make a hydrogen bond to camphor. Here, solvation of this cavity is calculated to be unfavorable by +15.8 +/- 5.0 kJ/mol. The molecular dynamics simulations can thus distinguish a hydrated interfacial cavity from an empty one. They also provide support for the notion that protein-ligand complexes can accommodate empty interfacial cavities and that such cavities are likely to be unhydrated unless more than one hydrogen bond can be made to a water molecule in the cavity.     

Incorporating replacement free energy of binding‐site waters in molecular docking

Binding‐site water molecules play a crucial role in protein‐ligand recognition, either being displaced upon ligand binding or forming water bridges to stabilize the complex. However, rigorously treating explicit binding‐site waters is challenging in molecular docking, which requires to fully sample ensembles of waters and to consider the free energy cost of replacing waters. Here, we describe a method to incorporate structural and energetic properties of binding‐site waters into molecular docking. We first developed a solvent property analysis (SPA) program to compute the replacement free energies of binding‐site water molecules by post‐processing molecular dynamics trajectories obtained from ligand‐free protein structure simulation in explicit water. Next, we implemented a distance‐dependent scoring term into DOCK scoring function to take account of the water replacement free energy cost upon ligand binding. We assessed this approach in protein targets containing important binding‐site waters, and we demonstrated that our approach is reliable in reproducing the crystal binding geometries of protein‐ligand‐water complexes, as well as moderately improving the ligand docking enrichment performance. In addition, SPA program (free available to academic users upon request) may be applied in identifying hot‐spot binding‐site residues and structure‐based lead optimization. Proteins 2014; 82:1765–1776. © 2014 Wiley Periodicals, Inc.     

Structural and dynamic roles of permanent water molecules in ligand molecular recognition by chicken liver bile acid binding protein

Chicken liver bile acid binding protein (cL-BABP) crystallizes with water molecules in its binding site. To obtain insights on the role of internal water, we performed two 100 ns molecular dynamics (MD) simulations in explicit solvent for cL-BABP, as apo form and as a complex with two molecules of cholic acid, and analyzed in detail the dynamics properties of all water molecules. The diffusion coefficients of the more persistent internal water molecules are significantly different from the bulk, but similar between the two protein forms. A different number of molecules and a different organization are observed for apo- and holo-cL-BABP. Most water molecules identified in the binding site of the apo-crystal diffuse to the bulk during the simulation. In contrast, almost all the internal waters of the holo-crystal maintain the same interactions with internal sidechains and ligands, which suggests they have a relevant role in protein-ligand molecular recognition. Only in the presence of these water molecules we were able to reproduce, by a classical molecular docking approach, the structure of the complex cL-BABP::cholic acid with a low ligand root mean square deviation (RMSD) with respect to its reference positioning. Literature data reported a conserved pattern of hydrogen bonds between a single water molecule and three amino acid residues of the binding site in a series of crystallized FABP. In cL-BABP, the interactions between this conserved water molecule and the three residues are present in the crystal of both apo- and holo-cL-BABP but are lost immediately after the start of molecular dynamics. Copyright (c) 2008 John Wiley & Sons, Ltd.     

Automatic identification and representation of protein binding sites for molecular docking.

Molecular docking is a popular way to screen for novel drug compounds. The method involves aligning small molecules to a protein structure and estimating their binding affinity. To do this rapidly for tens of thousands of molecules requires an effective representation of the binding region of the target protein. This paper presents an algorithm for representing a protein's binding site in a way that is specifically suited to molecular docking applications. Initially the protein's surface is coated with a collection of molecular fragments that could potentially interact with the protein. Each fragment, or probe, serves as a potential alignment point for atoms in a ligand, and is scored to represent that probe's affinity for the protein. Probes are then clustered by accumulating their affinities, where high affinity clusters are identified as being the "stickiest" portions of the protein surface. The stickiest cluster is used as a computational binding "pocket" for docking. This method of site identification was tested on a number of ligand-protein complexes; in each case the pocket constructed by the algorithm coincided with the known ligand binding site. Successful docking experiments demonstrated the effectiveness of the probe representation.     

Molecular mimicry of substrate oxygen atoms by water molecules in the beta-amylase active site

Soybean beta-amylase (EC 3.2.1.2) has been crystallized both free and complexed with a variety of ligands. Four water molecules in the free-enzyme catalytic cleft form a multihydrogen-bond network with eight strategic residues involved in enzyme-ligand hydrogen bonds. We show here that the positions of these four water molecules are coincident with the positions of four potential oxygen atoms of the ligands within the complex. Some of these waters are displaced from the active site when the ligands bind to the enzyme. How many are displaced depends on the shape of the ligand. This means that when one of the four positions is not occupied by a ligand oxygen atom, the corresponding water remains. We studied the functional/structural role of these four waters and conclude that their presence means that the conformation of the eight side chains is fixed in all situations (free or complexed enzyme) and preserved from unwanted or forbidden conformational changes that could hamper the catalytic mechanism. The water structure at the active pocket of beta-amylase is therefore essential for providing the ligand recognition process with plasticity. It does not affect the protein active-site geometry and preserves the overall hydrogen-bonding network, irrespective of which ligand is bound to the enzyme. We also investigated whether other enzymes showed a similar role for water. Finally, we discuss the potential use of these results for predicting whether water molecules can mimic ligand atoms in the active center.