Protein-based virtual screening of chemical databases. 1. Evaluation of different docking/scoring combinations

A practical approach to docking of zinc metalloproteinase inhibitors

Forty zinc-dependent metalloproteinase/ligand complexes with known crystal structures were re-docked using five docking/scoring approaches (DOCK, FlexX, DrugScore, GOLD, and AutoDock). Correct geometry of the coordination bonds between the ligand's zinc binding group (ZBG) and the catalytic zinc is important for docking accuracy and scoring reliability. More than 75% of docked poses with RMSD less than 2A were found to have appropriate ZBG binding, but for poor ZBG binding, about 95% of poses failed to dock correctly. Elimination of poses with inappropriate zinc binding resulted in better binding energy predictions that were further improved by dividing the ligands into subsets according to the ZBG (carboxylates, hydroxamates, and phosphorus containing groups). After a subset re-scoring using the regression functions obtained for individual subsets, DrugScore was able to explain 77% and the consensus scoring scheme X-CSCORE even 88% of variance in binding energies. The approach combining ZBG-based pose selection and subset re-scoring improved the hit rate in virtual screening for metalloproteinase inhibitors for all tested methods by 4-16%.

Novel technologies for virtual screening

There are several methods for virtual screening of databases of small organic compounds to find tight binders to a given protein target. Recent reviews in Drug Discovery Today have concentrated on screening by docking and by pharmacophore searching. Here, we complement these reviews by focusing on virtual screening methods that are based on analyzing ligand similarity on a structural level. Specifically, we concentrate on methods that exploit structural properties of the complete ligand molecules, as opposed to using just partial structural templates, such as pharmacophores. The in silico procedure of virtual screening (VS) and its relationship to the experimental procedure, HTS, is discussed, new developments in the field are summarized and perspectives on future research are offered.

A structure-based strategy to identify new molecular scaffolds targeting the bacterial ribosomal A-site

The need for novel antibiotics is widely recognized. A well validated target of antibiotics is the bacterial ribosome. Recent X-ray structures of the ribosome bound to antibiotics have shed new light on the binding sites of these antibiotics, providing fresh impetus for structure-based strategies aiming at identifying new ribosomal ligands. In that respect, the ribosomal decoding region of the aminoacyl-tRNA acceptor site (A-site) is of particular interest because oligonucleotide model systems of this site are available for crystallography, NMR and compound binding assays. This work presents how these different resources can be combined in a hierarchical screening strategy which has led to the identification of new A-site ligands. The approach exploits an X-ray structure of the A-site against which large and diverse libraries of compounds were computationally docked. The complementarity of the compounds to the A-site was assessed using a scoring function specifically calibrated for RNA targets. Starting from approximately 1 million compounds, the computational selection of candidate ligands allowed us to focus the experimental work on 129 compounds, 34 of which showed affinity for the A-site in a FRET-based binding assay. NMR experiments confirmed binding to the A-site for some compounds. For the most potent compound in the FRET assay, a tentative binding mode is suggested, which is compatible with the NMR data and the limited SAR in this series. Overall, the results validate the screening strategy.

HierVLS hierarchical docking protocol for virtual ligand screening of large-molecule databases

To provide practical means for rapidly scanning the extensive experimental combinatorial chemistry libraries now available for high-throughput screening (HTS), it is essential to establish computational virtual ligand screening (VLS) techniques to rapidly identify out of a large library all active compounds against a particular protein target. Toward this goal we developed HierVLS, a fast hierarchical docking approach that starts with a coarse grain conformational search over a large number of configurations filtered with a fast but crude energy function, followed by a succession of finer grain levels, using successively more accurate but more expensive descriptions of the ligand-protein-solvent interactions to filter successively fewer cases. The final step of this procedure optimizes one configuration of the ligand in the protein site using our most accurate energy expression and description of the solvent, which would be impractical for all conformations and sites sampled in the coarse level. HierVLS is based on the HierDock approach, but rather than allowing an hour or more to determine the best binding site and energy for each ligands (as in HierDock), we have adapted our procedure so that it can lead to reliable results while using only 4 min (866 MHz Pentium III processor) per ligand. To validate the accuracy for HierVLS to predict the experimentally observed binding conformation, we considered 37 cocrystal structures comprising 11 target proteins. We find that HierVLS identifies the correct binding mode for all 37 cocrystals. In addition, the calculated binding energies correlate well with available experimental binding constants. To validate how well HierVLS can identify the correct ligand in an extensive library of decoys, we considered a library of over 10 000 molecules. HierVLS identifies 26 out of the 37 cases in the top 2% ranked by binding affinity among the 10 037 molecules. The failures result from either metal-containing sites on the protein or water-mediated ligand-protein interactions, which we anticipate can be solved within the constraints of practical VLS. We then applied HierVLS to screen a 55000-compound virtual library against the target protein-tyrosine phosphatase 1B (ptp1b). The top 250 compounds by binding affinity included all six ptp1b cocrystal ligands added to the library plus three other experimentally confirmed binders. The best (top 1) binder is an experimentally confirmed positive. We conclude that HierVLS is useful for selecting leads for a particular target out of large combinatorial databases.

Multiple active site corrections for docking and virtual screening

Several docking programs are now available that can reproduce the bound conformation of a ligand in an active site, for a wide variety of experimentally determined complexes. However, these programs generally perform less well at ranking multiple possible ligands in one site. Since accurate identification of potential ligands is a prerequisite for many aspects of structure-based drug design, this is a serious limitation. We have tested the ability of two docking programs, FlexX and Gold, to match ligands and active sites for multiple complexes. We show that none of the docking scores from either program are able to match consistently ligands and active sites in our tests. We propose a simple statistical correction, the multiple active site correction (MASC), which greatly ameliorates this problem. We have also tested the correction method against an extended set of 63 cocrystals and in a virtual screening experiment. In all cases, MASC significantly improves the results of the docking experiments.

Rational design of an indolebutanoic acid derivative as a novel aldose reductase inhibitor based on docking and 3D QSAR studies of phenethylamine derivatives

A series of 45 phenethylamine derivatives were synthesized and evaluated for their inhibitory activity against pig kidney aldose reductase (ALR2, EC 1.1.1.21). Their IC(50) values ranged from 400 microM to 24 microM. The binding modes of compounds at the active site of ALR2 were examined using flexible docking. The results indicated that phenethylamine derivatives nicely fit into the active pocket of ALR2 by forming various hydrogen bonding and hydrophobic interactions. 3D-QSAR analysis was also conducted using FlexX-docked alignment of the compounds. The best prediction was obtained by CoMSIA combined with hydrophobic and hydrogen bond donor/acceptor field (q(2) = 0.557, r(2) = 0.934). A new derivative, 4-oxo-4-(4-hydroxyindole)butanoic acid, was designed, taking into account the CoMSIA field and the binding mode derived by FlexX docking. This rationally designed compound exhibits an ALR2 inhibition with an IC(50) value of 7.4 microM, which compares favorably to that of a well-known ALR2 inhibitor, tolrestat (IC(50) = 16 microM) and represents a potency approximately 240-fold higher than that of an original phenethylamine lead compound, YUA001.

Improving structure-based virtual screening by multivariate analysis of scoring data

Three different multivariate statistical methods, PLS discriminant analysis, rule-based methods, and Bayesian classification, have been applied to multidimensional scoring data from four different target proteins: estrogen receptor alpha (ERalpha), matrix metalloprotease 3 (MMP3), factor Xa (fXa), and acetylcholine esterase (AChE). The purpose was to build classifiers able to discriminate between active and inactive compounds, given a structure-based virtual screen. Seven different scoring functions were used to generate the scoring matrices. The classifiers were compared to classical consensus scoring and single scoring functions. The classifiers show a superior performance, with rule-based methods being most effective. The precision of correctly predicting an active compound is about 90% for three of the targets and about 25% for acetylcholine esterase. On the basis of these results, a new two-stage approach is suggested for structure-based virtual screening where limited activity information is available.

Prediction of Ligand Binding Affinity and Orientation of Xenoestrogens to the Estrogen Receptor by Molecular Dynamics Simulations and the Linear Interaction Energy Method

Exposure to environmental estrogens has been proposed as a risk factor for disruption of reproductive development and tumorigenesis of humans and wildlife (McLachlan, J. A.; Korach, K. S.; Newbold, R. R.; Degen, G. H. Diethylstilbestrol and other estrogens in the environment. Fundam. Appl. Toxicol. 1984, 4, 686-691). In recent years, many structurally diverse environmental compounds have been identified as estrogens. A reliable computational method for determining estrogen receptor (ER) binding affinity is of great value for the prediction of estrogenic activity of such compounds and their metabolites. In the presented study, a computational model was developed for prediction of binding affinities of ligands to the ERalpha isoform, using MD simulations in combination with the linear interaction energy (LIE) approach. The linear interaction energy approximation was first described by Aqvist et al. (Aqvist, J.; Medina, C.; Samuelsson, J. E. A new method for predicting binding affinity in computer-aided drug design. Protein Eng. 1994, 7, 385-391) and relies on the assumption that the binding free energy (DeltaG) depends linearly on changes in the van der Waals and electrostatic energy of the system. In the present study, MD simulations of ligands in the ERalpha ligand binding domain (LBD) (Shiau, A. K.; Barstad, D.; Loria, P. M.; Cheng, L.; Kushner, P. J.; Agard, D. A.; Greene, G. L. The structural basis of estrogen receptor/coactivator recognition and the antagonism of this interaction by tamoxifen. Cell 1998, 95, 927-937), as well as ligands free in water, were carried out using the Amber 6.0 force field (http://amber.scripps.edu/). Contrary to previous LIE methods, we took into account every possible orientation of the ligands in the LBD and weighted the contribution of each orientation to the total binding affinity according to a Boltzman distribution. The training set (n = 19) contained estradiol (E2), the synthetic estrogens diethylstilbestrol (DES) and 11beta-chloroethylestradiol (E2-Cl), 16alpha-hydroxy-E2 (estriol, EST), the phytoestrogens genistein (GEN), 8-prenylnaringenin (8PN), and zearalenon (ZEA), four derivatives of benz[a]antracene-3,9-diol, and eight estrogenic monohydroxylated PAH metabolites. We obtained an excellent linear correlation (r(2) = 0.94) between experimental (competitive ER binding assay) and calculated binding energies, with K(d) values ranging from 0.15 mM to 30 pM, a 5 000 000-fold difference in binding affinity. Subsequently, a test set (n = 12) was used to examine the predictive value of our model. This set consisted of the synthetic estrogen 5,11-cis-diethyl-5,6,11,12-tetrahydrochrysene-2,8-diol (THC), daidzein (DAI), equol (EQU) and apigenin (API), chlordecone (KEP), progesterone (PRG), several mono- and dihydroxylated PAH metabolites, and two brominated biphenyls. The predicted binding affinities of these estrogenic compounds were in very good agreement with the experimental values (average deviation of 0.61 +/- 0.4 kcal/mol). In conclusion, our LIE model provides a very good method for prediction of absolute ligand binding affinities, as well as binding orientation of ligands.

Comparing performance of computational tools for combinatorial library design

In using computational tools for library design it is necessary to understand the performance and limitations of available methods. This letter reports systematic comparisons of applying ligand-based and structure-based tools across therapeutic project-derived data sets. Included are assessments of performance in real-world iterative design applications and the utility of target structural information. The results suggest that combining screening and target structure information is robust; further, a well-designed screening library can compensate for lacking structural information.

Performance of 3D-Database Molecular Docking Studies into Homology Models

The performance of docking studies into protein active sites constructed by homology model building was investigated using CDK2 and factor VIIa screening data sets. When the sequence identity between model and template near the binding site area is greater than approximately 50%, roughly 5 times more active compounds are identified than would be found randomly. This performance is comparable to docking to crystal structures.

Evaluation of Docking Performance: Comparative Data on Docking Algorithms

Docking molecules into their respective 3D macromolecular targets is a widely used method for lead optimization. However, the best known docking algorithms often fail to position the ligand in an orientation close to the experimental binding mode. It was reported recently that consensus scoring enhances the hit rates in a virtual screening experiment. This methodology focused on the top-ranked pose, with the underlying assumption that the orientation/conformation of the docked compound is the most accurate. In an effort to eliminate the scoring function bias, and assess the ability of the docking algorithms to provide solutions similar to the crystallographic modes, we investigated the most known docking programs and evaluated all of the resultant poses. We present the results of an extensive computational study in which five docking programs (FlexX, DOCK, GOLD, LigandFit, Glide) were investigated against 14 protein families (69 targets). Our findings show that some algorithms perform consistently better than others, and a correspondence between the nature of the active site and the best docking algorithm can be found.

Predicting protein–ligand binding affinities: a low scoring game?

We have investigated the performance of five well known scoring functions in predicting the binding affinities of a diverse set of 205 protein-ligand complexes with known experimental binding constants, and also on subsets of mutually similar complexes. We have found that the overall performance of the scoring functions on the diverse set is disappointing, with none of the functions achieving r(2) values above 0.32 on the whole dataset. Performance on the subsets was mixed, with four of the five functions predicting fairly well the binding affinities of 35 proteinases, but none of the functions producing any useful correlation on a set of 38 aspartic proteinases. We consider two algorithms for producing consensus scoring functions, one based on a linear combination of scores from the five individual functions and the other on averaging the rankings produced by the five functions. We find that both algorithms produce consensus functions that generally perform slightly better than the best individual scoring function on a given dataset.

Fluctuation analysis and accuracy of a large-scalein silico screen

Using a cascadic version of the stochastic tunneling method we perform an all-atom database screen over 186,000 flexible ligands of the NCI 3D database against the thymidine kinase receptor. By analyzing the errors in the binding energy we demonstrate how the cascadic technique is superior to conventional sequential docking techniques and how reliable results for the determination of the top-scoring ligands could be achieved. The substrate corresponding to the crystal structure used in the screen ranks in the upper 0.05% of the database, validating both docking methodology and the applicability of the scoring function to this substrate. Several high ranking ligands of the database display significant structural similarity with known substrates. A detailed analysis of the accuracy of the screening method is carried out, and its dependence on the flexibility of the ligand is quantified.

The Role of Virtual Screening in Computer Aided Structure-Based Drug Design

The pharmaceutical industry has embraced computational methods to improve the successful negotiation of hits and leads into drugs in the clinic. This review examines the current status of in silico screening methods and aspects of compound library design.

Molecular recognition and docking algorithms

Molecular docking is an invaluable tool in modern drug discovery. This review focuses on methodological developments relevant to the field of molecular docking. The forces important in molecular recognition are reviewed and followed by a discussion of how different scoring functions account for these forces. More recent applications of computational chemistry tools involve library design and database screening. Last, we summarize several critical methodological issues that must be addressed in future developments.

2D QSAR Consensus Prediction for High-Throughput Virtual Screening. An Application to COX-2 Inhibition Modeling and Screening of the NCI Database

Using classification (SOM, LVQ, Binary, Decision Tree) and regression algorithms (PLS, BRANN, k-NN, Linear), this paper details the building of eight 2D-QSAR models from a 266 COX-2 inhibitor training set. The predictive performances of these eight models were subsequently compared using an 88 COX-2 inhibitor test set. Each ligand is described by 52 2D descriptors expressed as van der Waals Surface Areas (P_VSA) and its COX-2 binding IC50. One of our best predictive models is the neural network model (BRANN), which is able to select a subset, from the 88 ligand test set, that contains 94% COX-2 active inhibitors (pIC50>7.5) and detects 71% of all the actives. We then introduce a QSAR consensus prediction protocol that is shown to be more predictive than any single QSAR model: our C3 consensus approach is able to select a subset from the 88 ligand test set that contains 94% active inhibitors and 83% of all the actives. The 2D QSAR consensus protocol was finally applied to the high-throughput virtual screening of the NCI database, containing 193,477 organic compounds.

A convenient method for the computer-aided molecular design of carborane containing compounds

Computer-aided molecular design (CAMD) of carborane containing compounds is of growing interest for scientists involved in boron neutron capture therapy (BNCT) and other pharmaceutical applications. However, the complex organo-metallic structures of carboranes pose difficulties in modeling and docking of these structures. This is the first report of a new strategy for modeling and docking of carborane containing molecules with the readily available software packages HyperChem, SYBYL and FlexX. It is intended as a guide for boron chemists interested in using CAMD of carborane containing agents for medical applications such as BNCT.

Detailed analysis of grid-based molecular docking: A case study of CDOCKER-A CHARMm-based MD docking algorithm

The influence of various factors on the accuracy of protein-ligand docking is examined. The factors investigated include the role of a grid representation of protein-ligand interactions, the initial ligand conformation and orientation, the sampling rate of the energy hyper-surface, and the final minimization. A representative docking method is used to study these factors, namely, CDOCKER, a molecular dynamics (MD) simulated-annealing-based algorithm. A major emphasis in these studies is to compare the relative performance and accuracy of various grid-based approximations to explicit all-atom force field calculations. In these docking studies, the protein is kept rigid while the ligands are treated as fully flexible and a final minimization step is used to refine the docked poses. A docking success rate of 74% is observed when an explicit all-atom representation of the protein (full force field) is used, while a lower accuracy of 66-76% is observed for grid-based methods. All docking experiments considered a 41-member protein-ligand validation set. A significant improvement in accuracy (76 vs. 66%) for the grid-based docking is achieved if the explicit all-atom force field is used in a final minimization step to refine the docking poses. Statistical analysis shows that even lower-accuracy grid-based energy representations can be effectively used when followed with full force field minimization. The results of these grid-based protocols are statistically indistinguishable from the detailed atomic dockings and provide up to a sixfold reduction in computation time. For the test case examined here, improving the docking accuracy did not necessarily enhance the ability to estimate binding affinities using the docked structures.

FDS: flexible ligand and receptor docking with a continuum solvent model and soft-core energy function

The docking of flexible small molecule ligands to large flexible protein targets is addressed in this article using a two-stage simulation-based method. The methodology presented is a hybrid approach where the first component is a dock of the ligand to the protein binding site, based on deriving sets of simultaneously satisfied intermolecular hydrogen bonds using graph theory and a recursive distance geometry algorithm. The output structures are reduced in number by cluster analysis based on distance similarities. These structures are submitted to a modified Monte Carlo algorithm using the AMBER-AA molecular mechanics force field with the Generalized Born/Surface Area (GB/SA) continuum model. This solvent model is not only less expensive than an explicit representation, but also yields increased sampling. Sampling is also increased using a rotamer library to direct some of the protein side-chain movements along with large dihedral moves. Finally, a softening function for the nonbonded force field terms is used, enabling the potential energy function to be slowly turned on throughout the course of the simulation. The docking procedure is optimized, and the results are presented for a single complex of the arabinose binding protein. It was found that for a rigid receptor model, the X-ray binding geometry was reproduced and uniquely identified based on the associated potential energy. However, when side-chain flexibility was included, although the X-ray structure was identified, it was one of three possible binding geometries that were energetically indistinguishable. These results suggest that on relaxing the constraint on receptor flexibility, the docking energy hypersurface changes from being funnel-like to rugged. A further 14 complexes were then examined using the optimized protocol. For each complex the docking methodology was tested for a fully flexible ligand, both with and without protein side-chain flexibility. For the rigid protein docking, 13 out of the 15 test cases were able to find the experimental binding mode; this number was reduced to 11 for the flexible protein docking. However, of these 11, in the majority of cases the experimental binding mode was not uniquely identified, but was present in a cluster of low energy structures that were energetically indistinguishable. These results not only support the presence of a rugged docking energy hypersurface, but also suggest that it may be necessary to consider the possibility of more than one binding conformation during ligand optimization.

A web-based platform for virtual screening

A fully integrated, web-based, virtual screening platform has been developed to allow rapid virtual screening of large numbers of compounds. ORACLE is used to store information at all stages of the process. The system includes a large database of historical compounds from high throughput screenings (HTS) chemical suppliers, ATLAS, containing over 3.1 million unique compounds with their associated physiochemical properties (ClogP, MW, etc.). The database can be screened using a web-based interface to produce compound subsets for virtual screening or virtual library (VL) enumeration. In order to carry out the latter task within ORACLE a reaction data cartridge has been developed. Virtual libraries can be enumerated rapidly using the web-based interface to the cartridge. The compound subsets can be seamlessly submitted for virtual screening experiments, and the results can be viewed via another web-based interface allowing ad hoc querying of the virtual screening data stored in ORACLE.

Analysis and optimization of structure-based virtual screening protocols. 2. Examination of docked ligand orientation sampling methodology: mapping a pharmacophore for success

An important element of any structure-based virtual screening (SVS) technique is the method used to orient the ligands in the target active site. This has been a somewhat overlooked issue in recent SVS validation studies, with the assumption being made that the performance of an algorithm for a given set of orientation sampling settings will be representative for the general behavior of said technique. Here, we analyze five different SVS targets using a variety of sampling paradigms within the DOCK, GOLD and PROMETHEUS programs over a data set of approximately 10,000 noise compounds, combined with data sets containing multiple active compounds. These sets have been broken down by chemotype, with chemotype hit rate used to provide a measure of enrichment with a potentially improved relevance to real world SVS experiments. The variability in enrichment results produced by different sampling paradigms is illustrated, as is the utility of using pharmacophores to constrain sampling to regions that reflect known structural biology. The difference in results when comparing chemotype with compound hit rates is also highlighted.

Structural basis for contrasting activities of ribosome binding thiazole antibiotics

Thiostrepton and micrococcin inhibit protein synthesis by binding to the L11 binding domain (L11BD) of 23S ribosomal RNA. The two compounds are structurally related, yet they produce different effects on ribosomal RNA in footprinting experiments and on elongation factor-G (EF-G)-dependent GTP hydrolysis. Using NMR and an assay based on A1067 methylation by thiostrepton-resistance methyltransferase, we show that the related thiazoles, nosiheptide and siomycin, also bind to this region. The effect of all four antibiotics on EF-G-dependent GTP hydrolysis and EF-G-GDP-ribosome complex formation was studied. Our NMR and biochemical data demonstrate that thiostrepton, nosiheptide, and siomycin share a common profile, which differs from that of micrococcin. We have generated a three-dimensional (3D) model for the interaction of thiostrepton with L11BD RNA. The model rationalizes the differences between micrococcin and the thiostrepton-like antibiotics interacting with L11BD.

Receptor-drug association studies in the inhibition of the hematin aggregation process of malaria

Docking studies were performed to investigate the binding of several antimalarial compounds to the putative drug receptors involved in the hematin aggregation process. These studies reveal a binding profile that correlates with the complementarity of electrostatic potentials between the receptors and the active molecules. These results allow a possible explanation for the same molecular mechanism shown by 4-aminoquinolines, quinine, mefloquine, halofantrine and hydroxylated xanthones. The docking data presented in this work offer an interesting approach to the design of new molecules with potential antimalarial activity.

Conformational Changes of G Protein‐Coupled Receptors During Their Activation by Agonist Binding

The superfamily of G protein-coupled receptors (GPCRs) is the largest and most diverse group of transmembrane proteins involved in signal transduction. Many of the over 1000 human GPCRs represent important pharmaceutical targets. However, despite high interest in this receptor family, no high-resolution structure of a human GPCR has been resolved yet. This is mainly due to difficulties in obtaining large quantities of pure and active protein. Until now, only a high-resolution x-ray structure of an inactive state of bovine rhodopsin is available. Since no structure of an active state has been solved, information of the GPCR activation process can be gained only by biophysical techniques. In this review, we first describe what is known about the ground state of GPCRs to then address questions about the nature of the conformational changes taking place during receptor activation and the mechanism controlling the transition from the resting to the active state. Finally, we will also address the question to what extent information about the three-dimensional GPCR structure can be included into pharmaceutical drug design programs.

Nuclear hormone receptor targeted virtual screening

Virtual library screening (VLS) is emerging as a valuable drug lead discovery tool. ICM-VLS implementation of this technology was evaluated on a benchmark set of nuclear hormone receptors (NRs), an important therapeutic target family. Over 5000 structurally diverse compounds, including 78 known NR ligands, were screened against 18 crystal structures and one computer model of 10 NR ligand binding domains in their active or inactive states. The results confirm the ability of the VLS method to generate highly focused subsets of the input chemical library, enriched 33- to 100-fold for all but one receptor studied. However, receptor flexibility remains to be fully addressed, and the choice of the specific conformation used for screening may determine the success of the exercise. We observe that for a particular ligand VLS can often identify the correct target within the receptor family, although the technology is unable to reliably discriminate between the closely related receptor isoforms. Additionally, our results suggest that VLS may be applied successfully without an experimental structure of the receptor by using a homology model. These data represent a realistic snapshot of the state-of-the-art of NR-targeted VLS and define the recent progress and the remaining limitations of the technology.

BHB: a simple knowledge-based scoring function to improve the efficiency of database screening

A new knowledge-based scoring function was developed in this work to facilitate the rapid ranking of ligands in databases. The acronym of the method is BHB based on the descriptors it utilizes: buriedness, hydrogen bonding, and binding energy. Receptor buriedness is a measure of how well molecules occupy the binding pocket in comparison to known high-affinity ligands or, alternatively, whether they have contact with identified residues in the pocket. The possibility of hydrogen bond formation is checked for selected residues that are recognized as being important in the binding of known ligands. The approximate binding energy is calculated from the thermodynamic cycle using the optimized bound and free solvent conformations of the ligand-receptor system. The information necessary for the scoring function can ideally be gleaned from the 3D structure of the receptor-ligand complex. Alternatively, the descriptors can be derived from the 3D structure of the unbound receptor, provided this receptor has a known ligand that binds to the given site with nanomolar activity. We show that the new scoring functions provide up to 12 times improvement in enrichment compared to the popular commercial docking program GOLD.

Comparative evaluation of 11 scoring functions for molecular docking

Eleven popular scoring functions have been tested on 100 protein-ligand complexes to evaluate their abilities to reproduce experimentally determined structures and binding affinities. They include four scoring functions implemented in the LigFit module in Cerius2 (LigScore, PLP, PMF, and LUDI), four scoring functions implemented in the CScore module in SYBYL (F-Score, G-Score, D-Score, and ChemScore), the scoring function implemented in the AutoDock program, and two stand-alone scoring functions (DrugScore and X-Score). These scoring functions are not tested in the context of a particular docking program. Instead, conformational sampling and scoring are separated into two consecutive steps. First, an exhaustive conformational sampling is performed by using the AutoDock program to generate an ensemble of docked conformations for each ligand molecule. This conformational ensemble is required to cover the entire conformational space as much as possible rather than to focus on a few energy minima. Then, each scoring function is applied to score this conformational ensemble to see if it can identify the experimentally observed conformation from all of the other decoys. Among all of the scoring functions under test, six of them, i.e., PLP, F-Score, LigScore, DrugScore, LUDI, and X-Score, yield success rates higher than the AutoDock scoring function. The success rates of these six scoring functions range from 66% to 76% if using root-mean-square deviation < or =2.0 A as the criterion. Combining any two or three of these six scoring functions into a consensus scoring scheme further improves the success rate to nearly 80% or even higher. However, when applied to reproduce the experimentally determined binding affinities of the 100 protein-ligand complexes, only X-Score, PLP, DrugScore, and G-Score are able to give correlation coefficients over 0.50. All of the 11 scoring functions are further inspected by their abilities to construct a descriptive, funnel-shaped energy surface for protein-ligand complexation. The results indicate that X-Score and DrugScore perform better than the other ones at this aspect.

Calculation of the binding affinity of beta-secretase inhibitors using the linear interaction energy method

It has been shown that the rate-limiting step in the production of beta-amyloid peptide (Abeta) is the proteolytric cleavage of the membrane-bound beta-amyloid precursor protein (APP) by beta-secretase (BACE). Since the accumulation of Abeta has been implicated as one of the key events in the progression of Alzheimer's disease, BACE has become an important therapeutic target. Recently, two crystal structures of BACE cocrystallized with the inhibitors OM99-2 and OM00-3 were published by Tang and co-workers. In addition, the Ghosh group has published binding data on a series of inhibitors based on their initial lead, OM99-2. Using this set as a basis, we have developed a model for the binding affinity of these ligands to BACE using the linear interaction energy method. The best binding affinity model for the full set of ligands had a RMSD of 1.10 kcal/mol. The best model excluding the two charged ligands had a RMSD of 0.87 kcal/mol.

Hit and lead generation: beyond high-throughput screening

The identification of small-molecule modulators of protein function, and the process of transforming these into high-content lead series, are key activities in modern drug discovery. The decisions taken during this process have far-reaching consequences for success later in lead optimization and even more crucially in clinical development. Recently, there has been an increased focus on these activities due to escalating downstream costs resulting from high clinical failure rates. In addition, the vast emerging opportunities from efforts in functional genomics and proteomics demands a departure from the linear process of identification, evaluation and refinement activities towards a more integrated parallel process. This calls for flexible, fast and cost-effective strategies to meet the demands of producing high-content lead series with improved prospects for clinical success.

Surflex: fully automatic flexible molecular docking using a molecular similarity-based search engine

Surflex is a fully automatic flexible molecular docking algorithm that combines the scoring function from the Hammerhead docking system with a search engine that relies on a surface-based molecular similarity method as a means to rapidly generate suitable putative poses for molecular fragments. Results are presented evaluating reliability and accuracy of dockings compared with crystallographic experimental results on 81 protein/ligand pairs of substantial structural diversity. In over 80% of the complexes, Surflex's highest scoring docked pose was within 2.5 A root-mean-square deviation (rmsd), with over 90% of the complexes having one of the top ranked poses within 2.5 A rmsd. Results are also presented assessing Surflex's utility as a screening tool on two protein targets (thymidine kinase and estrogen receptor) using data sets on which competing methods were run. Performance of Surflex was significantly better, with true positive rates of greater than 80% at false positive rates of less than 1%. Docking time was roughly linear in number of rotatable bonds, beginning with a few seconds for rigid molecules and adding approximately 10 s per rotatable bond.

Binding site characteristics in structure-based virtual screening: evaluation of current docking tools

Two new docking programs FRED (OpenEye Scientific Software) and Glide (Schrödinger, Inc.) in combination with various scoring functions implemented in these programs have been evaluated against a variety of seven protein targets (cyclooxygenase-2, estrogen receptor, p38 MAP kinase, gyrase B, thrombin, gelatinase A, neuraminidase) in order to assess their accuracy in virtual screening. Sets of known inhibitors were added to and ranked relative to a random library of drug-like compounds. Performance was compared in terms of enrichment factors and CPU time consumption. Results and specific features of the two new tools are discussed and compared to previously published results using FlexX (Tripos, Inc.) as a docking engine. In addition, general criteria for the selection of docking algorithms and scoring functions based on binding-site characteristics of specific protein targets are proposed. Figure Enrichment factors obtained with FlexX, Glide and FRED docking engines in combination with different scoring functions for seven selected targets with highly variable binding sites

Protein-based virtual screening of chemical databases. II. Are homology models of G-Protein Coupled Receptors suitable targets?

The aim of the current study is to investigate whether homology models of G-Protein-Coupled Receptors (GPCRs) that are based on bovine rhodopsin are reliable enough to be used for virtual screening of chemical databases. Starting from the recently described 2.8 A-resolution X-ray structure of bovine rhodopsin, homology models of an "antagonist-bound" form of three human GPCRs (dopamine D3 receptor, muscarinic M1 receptor, vasopressin V1a receptor) were constructed. The homology models were used to screen three-dimensional databases using three different docking programs (Dock, FlexX, Gold) in combination with seven scoring functions (ChemScore, Dock, FlexX, Fresno, Gold, Pmf, Score). Rhodopsin-based homology models turned out to be suitable, indeed, for virtual screening since known antagonists seeded in the test databases could be distinguished from randomly chosen molecules. However, such models are not accurate enough for retrieving known agonists. To generate receptor models better suited for agonist screening, we developed a new knowledge- and pharmacophore-based modeling procedure that might partly simulate the conformational changes occurring in the active site during receptor activation. Receptor coordinates generated by this new procedure are now suitable for agonist screening. We thus propose two alternative strategies for the virtual screening of GPCR ligands, relying on a different set of receptor coordinates (antagonist-bound and agonist-bound states).

Virtual screening to enrich hit lists from high-throughput screening: a case study on small-molecule inhibitors of angiogenin

"Hit lists" generated by high-throughput screening (HTS) typically contain a large percentage of false positives, making follow-up assays necessary to distinguish active from inactive substances. Here we present a method for improving the accuracy of HTS hit lists by computationally based virtual screening (VS) of the corresponding chemical libraries and selecting hits by HTS/VS consensus. This approach was applied in a case study on the target-enzyme angiogenin, a potent inducer of angiogenesis. In conjunction with HTS of the National Cancer Institute Diversity Set and ChemBridge DIVERSet E (approximately 18,000 compounds total), VS was performed with two flexible library docking/scoring methods, DockVision/Ludi and GOLD. Analysis of the results reveals that dramatic enrichment of the HTS hit rate can be achieved by selecting compounds in consensus with one or both of the VS functions. For example, HTS hits ranked in the top 2% by GOLD included 42% of the true hits, but only 8% of the false positives; this represents a sixfold enrichment over the HTS hit rate. Notably, the HTS/VS method was effective in selecting out inhibitors with midmicromolar dissociation constants typical of leads commonly obtained in primary screens.

LigandFit: a novel method for the shape-directed rapid docking of ligands to protein active sites

We present a new shape-based method, LigandFit, for accurately docking ligands into protein active sites. The method employs a cavity detection algorithm for detecting invaginations in the protein as candidate active site regions. A shape comparison filter is combined with a Monte Carlo conformational search for generating ligand poses consistent with the active site shape. Candidate poses are minimized in the context of the active site using a grid-based method for evaluating protein-ligand interaction energies. Errors arising from grid interpolation are dramatically reduced using a new non-linear interpolation scheme. Results are presented for 19 diverse protein-ligand complexes. The method appears quite promising, reproducing the X-ray structure ligand pose within an RMS of 2A in 14 out of the 19 complexes. A high-throughput screening study applied to the thymidine kinase receptor is also presented in which LigandFit, when combined with LigScore, an internally developed scoring function, yields very good hit rates for a ligand pool seeded with known actives.

Consideration of molecular weight during compound selection in virtual target-based database screening

Virtual database screening allows for millions of chemical compounds to be computationally selected based on structural complimentary to known inhibitors or to a target binding site on a biological macromolecule. Compound selection in virtual database screening when targeting a biological macromolecule is typically based on the interaction energy between the chemical compound and the target macromolecule. In the present study it is shown that this approach is biased toward the selection of high molecular weight compounds due to the contribution of the compound size to the energy score. To account for molecular weight during energy based screening, we propose normalization strategies based on the total number of heavy atoms in the chemical compounds being screened. This approach is computationally efficient and produces molecular weight distributions of selected compounds that can be selected to be (1) lower than that of the original database used in the virtual screening, which may be desirable for selection of leadlike compounds or (2) similar to that of the original database, which may be desirable for the selection of drug-like compounds. By eliminating the bias in target-based database screening toward higher molecular weight compounds it is anticipated that the proposed procedure will enhance the success rate of computer-aided drug design.

Distilling the essential features of a protein surface for improving protein-ligand docking, scoring, and virtual screening

For the successful identification and docking of new ligands to a protein target by virtual screening, the essential features of the protein and ligand surfaces must be captured and distilled in an efficient representation. Since the running time for docking increases exponentially with the number of points representing the protein and each ligand candidate, it is important to place these points where the best interactions can be made between the protein and the ligand. This definition of favorable points of interaction can also guide protein structure-based ligand design, which typically focuses on which chemical groups provide the most energetically favorable contacts. In this paper, we present an alternative method of protein template and ligand interaction point design that identifies the most favorable points for making hydrophobic and hydrogen-bond interactions by using a knowledge base. The knowledge-based protein and ligand representations have been incorporated in version 2.0 of SLIDE and resulted in dockings closer to the crystal structure orientations when screening a set of 57 known thrombin and glutathione S-transferase (GST) ligands against the apo structures of these proteins. There was also improved scoring enrichment of the dockings, meaning better differentiation between the chemically diverse known ligands and a approximately 15,000-molecule dataset of randomly-chosen small organic molecules. This approach for identifying the most important points of interaction between proteins and their ligands can equally well be used in other docking and design techniques. While much recent effort has focused on improving scoring functions for protein-ligand docking, our results indicate that improving the representation of the chemistry of proteins and their ligands is another avenue that can lead to significant improvements in the identification, docking, and scoring of ligands.

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.

Structure-based virtual screening: an overview

Enormous advances in genomics have resulted in a large increase in the number of potential therapeutic targets that are available for investigation. This growth in potential targets has increased the demand for reliable target validation, as well as technologies that can identify rapidly several quality lead candidates. Virtual screening, and in particular receptor-based virtual screening, has emerged as a reliable, inexpensive method for identifying leads. Although still an evolving method, advances in computational techniques have enabled virtual screening to have a positive impact on the discovery process. Here, the current strengths and weaknesses of the technology are discussed, and emphasis is placed on aspects of the work-flow of a virtual screening campaign, from preparation through to post-screening analysis.

Structure activity relationship by NMR and by computer: a comparative study

There has recently been considerable interest in using NMR spectroscopy to identify ligand binding sites of macromolecules. In particular, a modular approach has been put forward by Fesik et al. (Shuker, S. B.; Hajduk, P. J.; Meadows, R. P.; Fesik, S. W. Science 1996, 274, 1531-1534) in which small ligands that bind to a particular target are identified in a first round of screening and subsequently linked together to form ligands of higher affinity. Similar strategies have also been proposed for in silico drug design, where the binding sites of small chemical groups are identified, and complete ligands are subsequently assembled from different groups that have favorable interactions with the macromolecular target. In this paper, we compare experimental and computational results on a selected target (FKBP12). The binding sites of three small ligands ((2S)1-acetylprolinemethylester, 1-formylpiperidine, 1-piperidinecarboxamide) in FKBP12 were identified independently by NMR and by computational methods. The subsequent comparison of the experimental and computational data showed that the computational method identified and ranked favorably ligand positions that satisfy the experimental NOE constraints.

Double evaluations of chemicals using a cocktail of fused recombinant receptors

Some chemicals have multipotential as endocrine-disrupting chemicals (EDCs). For example, some chemicals act as both estrogens and antiandrogens. Numbers of such chemicals should be evaluated from many aspects; however, labor and expenses are generally limited. We have developed two expression systems for the wild type of human estrogen receptor alpha and the wild type of human androgen receptor fused with a maltose binding protein. They are soluble and have binding activities. They showed dose-responses to natural hormones and well-known potential EDCs. After we established each assay condition for a competitive binding assay using each receptor, we found that two assay systems can be carried out simultaneously under limited and harmonized conditions. Under harmonized conditions using a cocktail of two types of receptors, we could estimate natural hormones and potential EDCs. Interference between two assay systems was not observed under these conditions. We believe that some competitive binding assays can be carried out using a cocktail of receptors at the same time if interference among different assay systems can be avoided by choosing ideal conditions.