Comparative evaluation of eight docking tools for docking and virtual screening accuracy

SODOCK: swarm optimization for highly flexible protein-ligand docking

Protein-ligand docking can be formulated as a parameter optimization problem associated with an accurate scoring function, which aims to identify the translation, orientation, and conformation of a docked ligand with the lowest energy. The parameter optimization problem for highly flexible ligands with many rotatable bonds is more difficult than that for less flexible ligands using genetic algorithm (GA)-based approaches, due to the large numbers of parameters and high correlations among these parameters. This investigation presents a novel optimization algorithm SODOCK based on particle swarm optimization (PSO) for solving flexible protein-ligand docking problems. To improve efficiency and robustness of PSO, an efficient local search strategy is incorporated into SODOCK. The implementation of SODOCK adopts the environment and energy function of AutoDock 3.05. Computer simulation results reveal that SODOCK is superior to the Lamarckian genetic algorithm (LGA) of AutoDock, in terms of convergence performance, robustness, and obtained energy, especially for highly flexible ligands. The results also reveal that PSO is more suitable than the conventional GA in dealing with flexible docking problems with high correlations among parameters. This investigation also compared SODOCK with four state-of-the-art docking methods, namely GOLD 1.2, DOCK 4.0, FlexX 1.8, and LGA of AutoDock 3.05. SODOCK obtained the smallest RMSD in 19 of 37 cases. The average 2.29 A of the 37 RMSD values of SODOCK was better than those of other docking programs, which were all above 3.0 A.

The consequences of scoring docked ligand conformations using free energy correlations

Ligands from a set of 19 protein-ligand complexes were re-docked with AutoDock, GOLD and FlexX using the scoring algorithms native to these programs supplemented by analysis using the HINT free energy force field. A HINT scoring function was calibrated for this data set using a simple linear regression of total HINT score for crystal-structure complexes vs. measured free energy of binding. This function had an r(2) of 0.84 and a standard error of +/-0.42 kcal mol(-1). The free energies of binding were calculated for the best poses using the AutoDock, GOLD and FlexX scoring functions. The AutoDock and GoldScore algorithms estimated more than half of the binding free energies within the reported calibration standard errors for these functions, while that of FlexX did not. In contrast, the calibrated HINT scoring function identified optimized poses with standard errors near +/-0.5 kcal mol(-1). When the metric of success is minimum RMSD (vs. crystallographic coordinates) the three docking programs were more successful, with mean RMSDs for the top-ranking poses in the 19 complexes of 3.38, 2.52 and 2.62 A for AutoDock, GOLD and FlexX, respectively. Two key observations in this study have general relevance for computational medicinal chemistry: first, while optimizing RMSD with docking score functions is clearly of value, these functions may be less well optimized for free energy of binding, which has broader applicability in virtual screening and drug discovery than RMSD; second, scoring functions uniquely calibrated for the data set or sets under study should nearly always be preferable to universal scoring functions. Due to these advantages, the poses selected by the HINT score also required less post-docking structure optimization to produce usable molecular models. Most of these features may be achievable with other scoring functions.

Benchmarking sets for molecular docking

Ligand enrichment among top-ranking hits is a key metric of molecular docking. To avoid bias, decoys should resemble ligands physically, so that enrichment is not simply a separation of gross features, yet be chemically distinct from them, so that they are unlikely to be binders. We have assembled a directory of useful decoys (DUD), with 2950 ligands for 40 different targets. Every ligand has 36 decoy molecules that are physically similar but topologically distinct, leading to a database of 98,266 compounds. For most targets, enrichment was at least half a log better with uncorrected databases such as the MDDR than with DUD, evidence of bias in the former. These calculations also allowed 40x40 cross-docking, where the enrichments of each ligand set could be compared for all 40 targets, enabling a specificity metric for the docking screens. DUD is freely available online as a benchmarking set for docking at http://blaster.docking.org/dud/.

A Common Reference Framework for Analyzing/Comparing Proteins and Ligands. Fingerprints for Ligands And Proteins (FLAP): Theory and Application

A fast new algorithm (Fingerprints for Ligands And Proteins or FLAP) able to describe small molecules and protein structures using a common reference framework of four-point pharmacophore fingerprints and a molecular-cavity shape is described in detail. The procedure starts by using the GRID force field to calculate molecular interaction fields, which are then used to identify particular target locations where an energetic interaction with small molecular features would be very favorable. The target points thus calculated are then used by FLAP to build all possible four-point pharmacophores present in the given target site. A related approach can be applied to small molecules, using directly the GRID atom types to identify pharmacophoric features, and this complementary description of the target and ligand then leads to several novel applications. FLAP can be used for selectivity studies or similarity analyses in order to compare macromolecules without superposing them. Protein families can be compared and clustered into target classes, without bias from previous knowledge and without requiring protein superposition, alignment, or knowledge-based comparison. FLAP can be used effectively for ligand-based virtual screening and structure-based virtual screening, with the pharmacophore molecular recognition. Finally, the new method can calculate descriptors for chemometric analysis and can initiate a docking procedure. This paper presents the background to the new procedure and includes case studies illustrating several relevant applications of the new approach.

Optimizing Fragment and Scaffold Docking by Use of Molecular Interaction Fingerprints

Protein-ligand interaction fingerprints have been used to postprocess docking poses of three ligand data sets: a set of 40 low-molecular-weight compounds from the Protein Data Bank, a collection of 40 scaffolds from pharmaceutically relevant protein ligands, and a database of 19 scaffolds extracted from true cdk2 inhibitors seeded in 2230 scaffold decoys. Four popular docking tools (FlexX, Glide, Gold, and Surflex) were used to generate poses for ligands of the three data sets. In all cases, scoring by the similarity of interaction fingerprints to a given reference was statistically superior to conventional scoring functions in posing low-molecular-weight fragments, predicting protein-bound scaffold coordinates according to the known binding mode of related ligands, and screening a scaffold library to enrich a hit list in true cdk2-targeted scaffolds.

Progressive Docking: A Hybrid QSAR/Docking Approach for Accelerating In Silico High Throughput Screening

A combination of protein-ligand docking and ligand-based QSAR approaches has been elaborated, aiming to speed-up the process of virtual screening. In particular, this approach utilizes docking scores generated for already processed compounds to build predictive QSAR models that, in turn, assess hypothetical target binding affinities for yet undocked entries. The "progressive docking" has been tested on drug-like substances from the NCI database that have been docked into several unrelated targets, including human sex hormone binding globulin (SHBG), carbonic anhydrase, corticosteroid-binding globulin, SARS 3C-like protease, and HIV1 reverse transcriptase. We demonstrate that progressive docking can reduce the amount of computations 1.2- to 2.6-fold (when compared to traditional docking), while maintaining 80-99% hit recovery rates. This progressive-docking procedure, therefore, substantially accelerates high throughput screening, especially when using high accuracy (slower) docking approaches and large-sized datasets, and has allowed us to identify several novel potent nonsteroidal SHBG ligands.

Computational and experimental approaches for assessing the interactions between the model calycin β-lactoglobulin and two antibacterial fluoroquinolones

Norfloxacin and levofloxacin, two fluoroquinolones of different bulk, rigidity and hydrophobicity taken as model ligands, were docked to one apo and two holo crystallographic structures of bovine beta-lactoglobulin (BLG) using different computational approaches. BLG is a member of the lipocalin superfamily. Lipocalins show a typical b-barrel structure encompassing an internal cavity where small hydrophobic molecules are usually bound. Our studies allowed the identification of two putative binding sites in addition to the calyx. The rigid docking approximation resulted in strong repulsive forces when the ligands were docked into the calyx of the apo form. On the contrary, hindrance was not experienced in flexible docking protocols whether on the apo or on the holo BLG forms, due to allowance for side chain rearrangement. K(i) between 10(-7) and 10(-6) M were estimated for norfloxacin at pH 7.4, smaller than 10(-5) M for levofloxacin. Spectroscopic and electrophoretic techniques experimentally validated the occurrence of an interaction between norfloxacin and BLG. Changes in chemical shift and dynamic parameters were observed between the (19)F NMR spectra of the complex and of the ligand. A K(i) (ca 10(-7) M) comparable with the docking results was estimated through a NMR relaxation titration. Stabilization against unfolding was demonstrated by denaturant gradient gel electrophoresis on the complex versus apo BLG. NMR experimental evidence points to a very loose interaction for ofloxacin, the racemic mixture containing levofloxacin. Furthermore, we were able to calculate in silico K(i)'s comparable to the published experimental values for the complexes of palmitic and retinoic acid with BLG.

Critical assessment of the automated AutoDock as a new docking tool for virtual screening

A major problem in virtual screening concerns the accuracy of the binding free energy between a target protein and a putative ligand. Here we report an example supporting the outperformance of the AutoDock scoring function in virtual screening in comparison to the other popular docking programs. The original AutoDock program is in itself inefficient to be used in virtual screening because the grids of interaction energy have to be calculated for each putative ligand in chemical database. However, the automation of the AutoDock program with the potential grids defined in common for all putative ligands leads to more than twofold increase in the speed of virtual database screening. The utility of the automated AutoDock in virtual screening is further demonstrated by identifying the actual inhibitors of various target enzymes in chemical databases with accuracy higher than the other docking tools including DOCK and FlexX. These results exemplify the usefulness of the automated AutoDock as a new promising tool in structure-based virtual screening.

InSilico-Guided Target Identification of a Scaffold-Focused Library: 1,3,5-Triazepan-2,6-diones as Novel Phospholipase A2 Inhibitors

A collection of 2150 druggable active sites from the Protein Data Bank was screened by high-throughput docking to identify putative targets for five representative molecules of a combinatorial library sharing a 1,3,5-triazepan-2,6-dione scaffold. Five targets were prioritized for experimental evaluation by computing enrichment in individual protein entries among the top 2% scoring targets. Out of the five proposed proteins, secreted phospholipase A2 (sPLA2) was shown to be a true target for a panel of 1,3,5-triazepan-2,6-diones which exhibited micromolar affinities toward two human sPLA2 members.

Contribution of Conformer Focusing to the Uncertainty in Predicting Free Energies for Protein−Ligand Binding

When a ligand binds to a protein, it is typically not in the lowest-energy conformation for the unbound ligand and there is also a loss of conformational degrees of freedom. The free-energy change for this "conformer focusing" is addressed here formally, and the associated errors with its estimation or neglect are considered in the context of scoring functions for protein-ligand docking and computation of absolute free energies of binding. Specific applications for inhibition of HIV-1 reverse transcriptase are reported. It is concluded that the uncertainties from this source alone are sufficient to preclude the viability of current docking methodology for rank-ordering of diverse compounds in high-throughput virtual screening.

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.

GFscore: a general nonlinear consensus scoring function for high-throughput docking

Most of the recent published works in the field of docking and scoring protein/ligand complexes have focused on ranking true positives resulting from a Virtual Library Screening (VLS) through the use of a specified or consensus linear scoring function. In this work, we present a methodology to speed up the High Throughput Screening (HTS) process, by allowing focused screens or for hitlist triaging when a prohibitively large number of hits is identified in the primary screen, where we have extended the principle of consensus scoring in a nonlinear neural network manner. This led us to introduce a nonlinear Generalist scoring Function, GFscore, which was trained to discriminate true positives from false positives in a data set of diverse chemical compounds. This original Generalist scoring Function is a combination of the five scoring functions found in the CScore package from Tripos Inc. GFscore eliminates up to 75% of molecules, with a confidence rate of 90%. The final result is a Hit Enrichment in the list of molecules to investigate during a research campaign for biological active compounds where the remaining 25% of molecules would be sent to in vitro screening experiments. GFscore is therefore a powerful tool for the biologist, saving both time and money.

Minimizing false positives in kinase virtual screens

In spite of recent improvements in docking and scoring methods, high false-positive rates remain a common issue in structure-based virtual screening. In this study, the distinctive features of false positives in kinase virtual screens were investigated. A series of retrospective virtual screens on kinase targets was performed on specifically designed test sets, each combining true ligands and experimentally confirmed inactive compounds. A systematic analysis of the docking poses generated for the top-ranking compounds highlighted key aspects differentiating true hits from false positives. The most recurring feature in the poses of false positives was the absence of certain key interactions known to be required for kinase binding. A systematic analysis of 444 crystal structures of ligand-bound kinases showed that at least two hydrogen bonds between the ligand and the backbone protein atoms in the kinase hinge region are present in 90% of the complexes, with very little variability across targets. Closer inspection showed that when the two hydrogen bonds are present, one of three preferred hinge-binding motifs is involved in 96.5% of the cases. Less than 10% of the false positives satisfied these two criteria in the minimized docking poses generated by our standard protocol. Ligand conformational artifacts were also shown to contribute to the occurrence of false positives in a number of cases. Application of this knowledge in the form of docking constraints and post-processing filters provided consistent improvements in virtual screening performance on all systems. The false-positive rates were significantly reduced and the enrichment factors increased by an average of twofold. On the basis of these results, a generalized two-step protocol for virtual screening on kinase targets is suggested.

In silico prediction of drug binding to CYP2D6: identification of a new metabolite of metoclopramide

Patients with cancer often take many different classes of drugs to treat the effects of their malignancy and the side effects of treatment, as well as their comorbidities. The potential for drug-drug interactions that may affect the efficacy of anticancer treatment is high, and a major source of such interactions is competition for the drug-metabolizing enzymes, cytochromes P450 (P450s). We have examined a series of 20 drugs commonly prescribed to cancer patients to look for potential interactions via CYP2D6. We used a homology model of CYP2D6, together with molecular docking techniques, to perform an in silico screen for binding to CYP2D6. Experimental IC50 values were determined for these compounds and compared with the model predictions to reveal a correlation with a regression coefficient of r2= 0.61. Importantly, the docked conformation of the commonly prescribed antiemetic metoclopramide predicted a new site of metabolism that was further investigated through in vitro analysis with recombinant CYP2D6. An aromatic N-hydroxy metabolite of metoclopramide, consistent with predictions from our modeling studies, was identified by high-performance liquid chromatography/mass spectrometry. This metabolite was found to represent a major product of metabolism in human liver microsomes, and CYP2D6 was identified as the main P450 isoform responsible for catalyzing its formation. In view of the prevalence of interindividual variation in the CYP2D6 genotype and phenotype, we suggest that those experiencing adverse reactions with metoclopramide, e.g., extrapyramidal syndrome, are likely to have a particular CYP2D6 genotype/phenotype. This warrants further investigation.

Effective handling of induced-fit motion in flexible docking

For structure-based drug design, where various ligand structures need to be docked to a target protein structure, a docking method that can handle conformational flexibility of not only the ligand, but also the protein, is indispensable. We have developed a simple and effective approach for dealing with the local induced-fit motion of the target protein, and implemented it in our docking tool, ADAM. Our approach efficiently combines the following two strategies: a vdW-offset grid in which the protein cavity is enlarged uniformly, and structure optimization allowing the motion of ligand and protein atoms. To examine the effectiveness of our approach, we performed docking validation studies, including redocking in 18 test cases and foreign-docking, in which various ligands from foreign crystal structures of complexes are docked into a target protein structure, in 22 cases (on five target proteins). With the original ADAM, the correct docking modes (RMSD < 2.0 A) were not present among the top 20 models in one case of redocking and four cases of foreign-docking. When the handling of induced-fit motion was implemented, the correct solutions were acquired in all 40 test cases. In foreign-docking on thymidine kinase, the correct docking modes were obtained as the top-ranked solutions for all 10 test ligands by our combinatorial approach, and this appears to be the best result ever reported with any docking tool. The results of docking validation have thus confirmed the effectiveness of our approach, which can provide reliable docking models even in the case of foreign-docking, where conformational change of the target protein cannot be ignored. We expect that this approach will contribute substantially to actual drug design, including virtual screening.

Binding mode of new (thio)hydantoin inhibitors of fatty acid amide hydrolase: comparison with two original compounds, OL-92 and JP104

Substituted (thio)hydantoins (2-thioxoimidazolidinones and imidazolidinediones) were reported as new potential reversible inhibitors of fatty acid amide hydrolase (FAAH). Their binding mode to FAAH was explored to rationalize their activity and give idea to design highly active inhibitors. Starting from the crystal structure of one of these molecules, docking studies provide us with rational basis for the design of new inhibitors within the thiohydantoin family.

TarFisDock: a web server for identifying drug targets with docking approach

TarFisDock is a web-based tool for automating the procedure of searching for small molecule-protein interactions over a large repertoire of protein structures. It offers PDTD (potential drug target database), a target database containing 698 protein structures covering 15 therapeutic areas and a reverse ligand-protein docking program. In contrast to conventional ligand-protein docking, reverse ligand-protein docking aims to seek potential protein targets by screening an appropriate protein database. The input file of this web server is the small molecule to be tested, in standard mol2 format; TarFisDock then searches for possible binding proteins for the given small molecule by use of a docking approach. The ligand-protein interaction energy terms of the program DOCK are adopted for ranking the proteins. To test the reliability of the TarFisDock server, we searched the PDTD for putative binding proteins for vitamin E and 4H-tamoxifen. The top 2 and 10% candidates of vitamin E binding proteins identified by TarFisDock respectively cover 30 and 50% of reported targets verified or implicated by experiments; and 30 and 50% of experimentally confirmed targets for 4H-tamoxifen appear amongst the top 2 and 5% of the TarFisDock predicted candidates, respectively. Therefore, TarFisDock may be a useful tool for target identification, mechanism study of old drugs and probes discovered from natural products. TarFisDock and PDTD are available at http://www.dddc.ac.cn/tarfisdock/.

Considerations in compound database preparation--"hidden" impact on virtual screening results

Structure-based virtual screening (SBVS) utilizing docking algorithms has become an essential tool in the drug discovery process, and significant progress has been made in successfully applying the technique to a wide range of receptor targets. In silico validation of virtual screening protocols before application to a receptor target using a corporate or commercially available compound collection is key to establishing a successful process. Ultimately, retrieval of a set of active compounds from a database of inactives is required, and the metric of enrichment (E) is habitually used to discern the quality of separation of the two. Numerous reports have addressed the performance of docking algorithms with regard to the quality of binding mode prediction and the issue of postprocessing "hit lists" of docked ligands. However, the impact of ligand database preprocessing has yet to be examined in the context of virtual screening and prioritization of compounds for biological evaluation. We provide an insight into the implications of cheminformatic preprocessing of a validation database of compounds where multiple protonated, tautomeric, stereochemical, and conformational states have been enumerated. Several commonly used methods for the generation of ligand conformations and conformational ensembles are examined, paired with an exhaustive rigid-body algorithm for the docking of different "multimeric" compound representations to the ligand binding site of the human estrogen receptor alpha. Chemgauss, a shapegaussian scoring function with intrinsic chemical knowledge, was combined with PLP as a consensus-scoring scheme to rank output from the docking protocol and enrichment rates calculated for each screen. The overheads of CPU consumption and the effect on relative database size (disk requirement) for each of the protocols employed are considered. Assessment of these parameters indicates that SBVS enrichments are highly dependent on the initial cheminformatic treatment(s) used in database construction. The interplay of SMILES representations, stereochemical information, protonation state enumeration, and ligand conformation ensembles are critical in achieving optimum enrichment rates in such screening.

Robust ligand-based modeling of the biological targets of known drugs

Systematic annotation of the primary targets of roughly 1000 known therapeutics reveals that over 700 of these modulate approximately 85 biological targets. We report the results of three analyses. In the first analysis, drug/drug similarities and target/target similarities were computed on the basis of three-dimensional ligand structures. Drug pairs sharing a target had significantly higher similarity than drug pairs sharing no target. Also, target pairs with no overlap in annotated drug specificity shared lower similarity than target pairs with increasing overlap. Two-way agglomerative clusterings of drugs and targets were consistent with known pharmacology and suggestive that side effects and drug-drug interactions might be revealed by modeling many targets. In the second analysis, we constructed and tested ligand-based models of 22 diverse targets in virtual screens using a background of screening molecules. Greater than 100-fold enrichment of cognate versus random molecules was observed in 20/22 cases. In the third analysis, selectivity of the models was tested using a background of drug molecules, with selectivity of greater than 80-fold observed in 17/22 cases. Predicted activities derived from crossing drugs against modeled targets identified a number of known side effects, drug specificities, and drug-drug interactions that have a rational basis in molecular structure.

Catalytic site prediction and virtual screening of cytochrome P450 2D6 substrates by consideration of water and rescoring in automated docking

Automated docking strategies successfully applied to binding mode predictions of ligands in Cyt P450 crystal structures in an earlier study (de Graaf et al. J. Med. Chem. 2005, 7, 2308-2318) were used for the catalytic site prediction (CSP) of 65 substrates in a CYP2D6 homology model. The consideration of water molecules at predicted positions in the active site and the rescoring of pooled docking poses from four different docking programs (AutoDock, FlexX, GOLD-Goldscore, and GOLD-Chemscore) with the SCORE scoring function enabled the successful prediction of experimentally reported sites of catalysis of more than 80% of the substrates. Three docking algorithms (FlexX, GOLD-Goldscore, and GOLD-Chemscore) were subsequently used in combination with six scoring functions (Chemscore, DOCK, FlexX, GOLD, PMF, and SCORE) to assess the ability of docking-based virtual screening methods to prioritize known CYP2D6 substrates seeded into a drug-like chemical database (in the absence and presence of active-site water molecules). Finally, the optimal docking strategy in terms of virtual screening accuracy, GOLD-Chemscore with the consideration of active-site water (60% of known substrates recovered in the top 5% of the ranked drug-like database), was verified experimentally; it was successfully used to identify high-affinity CYP2D6 ligands among a larger proprietary database.

Assessing the Discriminatory Power of Scoring Functions for Virtual Screening

The efficiency of scoring functions for hit identification is usually quantified in terms of enrichment factors and enrichment curves. Close inspection of simulated and real score distributions from virtual screening, however, suggests that 'analysis of variance' (ANOVA) is a more reliable method for assessing their performance. Using ANOVA to quantify the discriminatory power of scoring functions with respect to ligands, decoys, and a reproducible reference database has the potential to facilitate the advancement of scoring functions significantly.

Comparison of a homology model and the crystallographic structure of human 11β-hydroxysteroid dehydrogenase type 1 (11βHSD1) in a structure-based identification of inhibitors

Human 11beta-hydroxysteroid dehydrogenase type 1 (11betaHSD1) catalyzes the interconversion of cortisone into active cortisol. 11betaHSD1 inhibition is a tempting target for the treatment of a host of human disorders that might benefit from blockade of glucocorticoid action, such as obesity, metabolic syndrome, and diabetes type 2. Here, we report an in silico screening study aimed at identifying new selective inhibitors of human 11betaHSD1 enzyme. In the first step, homology modeling was employed to build the 3D structure of 11betaHSD1. Further, molecular docking was used to validate the predicted model by showing that it was able to discriminate between known 11betaHSD1 inhibitors or substrates and non-inhibitors. The homology model was found to reproduce closely the crystal structure that became publicly available in the final stages of this work. Finally, we carried out structure-based virtual screening experiments on both the homology model and the crystallographic structure with a database of 114,000 natural molecules. Among these, 15 molecules were consistently selected as inhibitors based on both the model and crystal structures of the enzyme, implying a good quality for the homology model. Among these putative 11betaHSD1 inhibitors, two were flavonone derivatives that have already been shown to be potent inhibitors of the enzyme.

Modelling of carbohydrate-aromatic interactions: ab initio energetics and force field performance

Aromatic amino acid residues are often present in carbohydrate-binding sites of proteins. These binding sites are characterized by a placement of a carbohydrate moiety in a stacking orientation to an aromatic ring. This arrangement is an example of CH/pi interactions. Ab initio interaction energies for 20 carbohydrate-aromatic complexes taken from 6 selected ultra-high resolution X-ray structures of glycosidases and carbohydrate-binding proteins were calculated. All interaction energies of a pyranose moiety with a side chain of an aromatic residue were calculated as attractive with interaction energy ranging from -2.8 to -12.3 kcal/mol as calculated at the MP2/6-311+G(d) level. Strong attractive interactions were observed for a wide range of orientations of carbohydrate and aromatic ring as present in selected X-ray structures. The most attractive interaction was associated with apparent combination of CH/pi interactions and classical H-bonds. The failure of Hartree-Fock method (interaction energies from +1.0 to -6.9 kcal/mol) can be explained by a dispersion nature of a majority of the studied complexes. We also present a comparison of interaction energies calculated at the MP2 level with those calculated using molecular mechanics force fields (OPLS, GROMOS, CSFF/CHARMM, CHEAT/CHARMM, Glycam/AMBER, MM2 and MM3). For a majority of force fields there was a strong correlation with MP2 values. RMSD between MP2 and force field values were 1.0 for CSFF/CHARMM, 1.2 for Glycam/AMBER, 1.2 for GROMOS, 1.3 for MM3, 1.4 for MM2, 1.5 for OPLS and to 2.3 for CHEAT/CHARMM (in kcal/mol). These results show that molecular mechanics approximates interaction energies very well and support an application of molecular mechanics methods in the area of glycochemistry and glycobiology.

In silico and in vitro screening for inhibition of cytochrome P450 CYP3A4 by comedications commonly used by patients with cancer

Cytochrome P450 3A4 (CYP3A4) is the major enzyme responsible for phase I drug metabolism of many anticancer agents. It is also a major route for metabolism of many drugs used by patients to treat the symptoms caused by cancer and its treatment as well as their other illnesses, for example, cardiovascular disease. To assess the ability to inhibit CYP3A4 of drugs most commonly used by our patients during cancer therapy, we have made in silico predictions based on the crystal structures of CYP3A4. From this set of 33 common comedicated drugs, 10 were predicted to be inhibitors of CYP3A4, with the antidiarrheal drug loperamide predicted to be the most potent. There was significant correlation (r(2) = 0.75-0.66) between predicted affinity and our measured IC(50) values, and loperamide was confirmed as a potent inhibitor (IC(50) of 0.050 +/- 0.006 microM). Active site docking studies predicted an orientation of loperamide consistent with formation of the major (N-demethylated) metabolite, where it interacts with the phenylalanine cluster and Arg-212 and Glu-374; experimental evidence for the latter interaction comes from the approximately 12-fold increase in K(M) for loperamide observed for the Glu-374-Gln mutant. The commonly prescribed drugs loperamide, amitriptyline, diltiazem, domperidone, lansoprazole, omeprazole, and simvastatin were identified by our in silico and in vitro screens as relatively potent inhibitors of CYP3A4 that have the potential to interact with cytotoxic agents to cause adverse effects, highlighting the likelihood of drug-drug interactions affecting chemotherapy treatment.

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.

Novel procedure for modeling ligand/receptor induced fit effects

We present a novel protein-ligand docking method that accurately accounts for both ligand and receptor flexibility by iteratively combining rigid receptor docking (Glide) with protein structure prediction (Prime) techniques. While traditional rigid-receptor docking methods are useful when the receptor structure does not change substantially upon ligand binding, success is limited when the protein must be "induced" into the correct binding conformation for a given ligand. We provide an in-depth description of our novel methodology and present results for 21 pharmaceutically relevant examples. Traditional rigid-receptor docking for these 21 cases yields an average RMSD of 5.5 A. The average ligand RMSD for docking to a flexible receptor for the 21 pairs is 1.4 A; the RMSD is < or =1.8 A for 18 of the cases. For the three cases with RMSDs greater than 1.8 A, the core of the ligand is properly docked and all key protein/ligand interactions are captured.

An evaluation of automated in silico ligand docking of amino acid ligands to Family C G-protein coupled receptors

Family C G-protein coupled receptors (GPCRs) consist of the metabotropic glutamate receptors (mGluRs), the calcium-sensing receptor (CaSR), the T1R taste receptors, the GABA(B) receptor, the V2R pheromone receptors, and several chemosensory receptors. A common feature of Family C receptors is the presence of an amino acid binding pocket. The objective of this study was to evaluate the ability of the automatic docking program FlexX to predict the favored amino acid ligand at several Family C GPCRs. The docking process was optimized using the crystal structure of mGluR1 and the 20 amino acids were docked into homology models of the CaSR, the 5.24 chemosensory receptor, and the GPRC6A amino acid receptor. Under optimized docking conditions, glutamate was docked in the binding pocket of mGluR1 with a root mean square deviation of 1.56 angstroms from the co-crystallized glutamate structure and was ranked as the best ligand with a significantly better FlexX score compared to all other amino acids. Ligand docking to a homology model of the 5.24 receptor gave generally correct predictions of the favored amino acids, while the results obtained with models of GPRC6A and the CaSR showed that some of the favored amino acids at these receptors were correctly predicted, while a few other top scoring amino acids appeared to be false positives. We conclude that with certain caveats, FlexX can be successfully used to predict preferred ligands at Family C GPCRs.

A Comparative Study on the Application of Hierarchical−Agglomerative Clustering Approaches to Organize Outputs of Reiterated Docking Runs

Reiterated runs of standard docking protocols usually provide a collection of possible binding modes rather than pinpoint a single solution. Usually, this ensemble is then ranked by means of an energy-based scoring function. However, since many degrees of approximation have to be introduced in the computation of the binding free energy, scoring functions cannot always rank the experimental pose among the top scorers. Cluster analysis might help to overcome this limit, provided that data clusterability has been earlier assessed. In this paper, first, we present a modified version of a test earlier developed by Hopkins to assess whether or not docking outputs show the natural tendency to be grouped in clusters. Then, we report the results of a comparative study on the application of different hierarchical-agglomerative cluster rules to partition docking outputs. The rule that was able to best manage the observed data was finally applied to the whole ensemble of poses collected from several docking tools. The combination of the average linkage rule with the cutting function developed by Sutcliffe and co-workers turned out to be an approach that meets all of the criteria required for a robust clustering protocol. Furthermore, a consensus clustering allowed us to identify the pose closest to the experimental one within a statistically significant cluster, whose number was always of few units.

Development and virtual screening of target libraries

The concomitant development of in silico screening technologies and of three-dimensional information on therapeutically relevant macromolecular targets makes it possible to navigate in the structural proteome and to identify targets fulfilling user-defined queries. This review illustrates some in-house recent advances in the development of target libraries and how they can be browsed to unravel chemogenomic information.

Structural e-bioinformatics and drug design

Nowadays the in silico scenario for drug design is totally dependent on structural biology and structural bioinformatics. A myriad of free bioinformatics applications and services have been posted on the web. This mini-review mentions web sites that are useful in structure-based drug design. The information is given in a logical manner, following the drug design process i.e. characterization of a protein target, modelling the protein using sequence homology, optimization of the protein structure and finally docking of small ligands into the active site.

Receptor systems: will mining the receptorome yield novel targets for pharmacotherapy?

We have recently defined the receptorome as 'that part of the proteome encoding receptors'. In this article, I provide a general overview of the members of the receptorome as well as methods used to screen the receptorome-both in silico and physically. Case histories of receptorome-based discovery efforts are then highlighted and the relevance of this approach to the discovery and validation of molecular targets for drug abuse treatment is emphasized.

Surrogate docking: structure-based virtual screening at high throughput speed

Structure-based screening using fully flexible docking is still too slow for large molecular libraries. High quality docking of a million molecule library can take days even on a cluster with hundreds of CPUs. This performance issue prohibits the use of fully flexible docking in the design of large combinatorial libraries. We have developed a fast structure-based screening method, which utilizes docking of a limited number of compounds to build a 2D QSAR model used to rapidly score the rest of the database. We compare here a model based on radial basis functions and a Bayesian categorization model. The number of compounds that need to be actually docked depends on the number of docking hits found. In our case studies reasonable quality models are built after docking of the number of molecules containing approximately 50 docking hits. The rest of the library is screened by the QSAR model. Optionally a fraction of the QSAR-prioritized library can be docked in order to find the true docking hits. The quality of the model only depends on the training set size - not on the size of the library to be screened. Therefore, for larger libraries the method yields higher gain in speed no change in performance. Prioritizing a large library with these models provides a significant enrichment with docking hits: it attains the values of approximately 13 and approximately 35 at the beginning of the score-sorted libraries in our two case studies: screening of the NCI collection and a combinatorial libraries on CDK2 kinase structure. With such enrichments, only a fraction of the database must actually be docked to find many of the true hits. The throughput of the method allows its use in screening of large compound collections and in the design of large combinatorial libraries. The strategy proposed has an important effect on efficiency but does not affect retrieval of actives, the latter being determined by the quality of the docking method itself.

Combining in Silico Tools and NMR Data To Validate Protein−Ligand Structural Models: Application to Matrix Metalloproteinases

A combination of in silico tools and experimental NMR data is proposed for relatively fast determination of protein-ligand structural models and demonstrated from known inhibitors of matrix metalloproteinases (MMP). The 15N 1H heteronuclear single quantum coherence (HSQC) spectral assignment and the 3D structure, either X-ray or NMR, are needed. In this method, the HSQC spectrum with or without the ligand is used to determine the interaction region of the ligand. Docking calculations are then performed to obtain a set of structural models. From the latter, the nuclear Overhauser effects (NOEs) between the ligand and the protein can be predicted. Guided by these predictions, a number of NOEs can be detected and assigned through a HSQC NOESY experiment. These data are used as structural restraints to reject/refine the initial structural models through further in silico work. For a test protein (MMP-12, human macrophage metalloelastase), a final structure of a protein-ligand adduct was obtained that matches well with the full structural determination. A number of structural predictions were then made for adducts of a similar protein (MMP-1, human fibroblast collagenase) with the same and different ligands. The quality of the final results depended on the type and number of experimental NOEs, but in all cases, a well-defined ligand conformation in the protein binding site was obtained. This protocol is proposed as a viable alternative to the many approaches described in the literature.

Application of Molecular Modeling to Analysis of Inhibition of Kinesin Motor Proteins of the BimC Subfamily by Monastrol and Related Compounds

Application of molecular modeling approaches has potential to contribute to rational drug design. These approaches may be especially useful when attempting to elucidate the structural features associated with novel drug targets. In this study, molecular docking and molecular dynamics were applied to studies of inhibition of the human motor protein denoted HsEg5 and other homologues in the BimC subfamily. These proteins are essential for mitosis, so compounds that inhibit their activity may have potential as anticancer therapeutics. The discovery of a small-molecule cell-permeable inhibitor, monastrol, has stimulated research in this area. Interestingly, monastrol is reported to inhibit the human and Xenopus forms of Eg5, but not those from Drosophila and Aspergillus. In this study, homology modeling was used to generate models of the Xenopus, Drosophila, and Aspergillus homologues, using the crystal structure of the human protein in complex with monastrol as a template. A series of known inhibitors was docked into each of the homologues, and the differences in binding energies were consistent with reported experimental data. Molecular dynamics revealed significant changes in the structure of the Aspergillus homologue that may contribute to its relative insensitivity to monastrol and related compounds.

Multiple ligand-binding modes in bacterial R67 dihydrofolate reductase

R67 dihydrofolate reductase (DHFR), a bacterial plasmid-encoded enzyme associated with resistance to the drug trimethoprim, shows neither sequence nor structural homology with the chromosomal DHFR. It presents a highly symmetrical toroidal structure, where four identical monomers contribute to the unique central active-site pore. Two reactants (dihydrofolate, DHF), two cofactors (NADPH) or one of each (R67*DHF*NADPH) can be found simultaneously within the active site, the last one being the reactive ternary complex. As the positioning of the ligands has proven elusive to empirical determination, we addressed the problem from a theoretical perspective. Several potential structures of the ternary complex were generated using the docking programs AutoDock and FlexX. The variability among the final poses, many of which conformed to experimental data, prompted us to perform a comparative scoring analysis and molecular dynamics simulations to assess the stability of the complexes. Analysis of ligand-ligand and ligand-protein interactions along the 4 ns trajectories of eight different structures allowed us to identify important inter-ligand contacts and key protein residues. Our results, combined with published empirical data, clearly suggest that multipe binding modes of the ligands are possible within R67 DHFR. While the pterin ring of DHF and the nicotinamide ring of NADPH assume a stacked endo-conformation at the centre of the pore, probably assisted by V66, Q67 and I68, the tails of the molecules extend towards opposite ends of the cavity, adopting multiple configurations in a solvent rich-environment where hydrogen-bond interactions with K32 and Y69 may play important roles.

A Flexible Docking Procedure for the Exploration of Peptide Binding Selectivity to Known Structures and Homology Models of PDZ Domains

PDZ domains are important scaffolding modules that typically bind to the C-termini of their interaction partners. Several structures of such complexes have been solved, revealing a conserved binding site in the PDZ domain and an extended conformation of the bound peptide. A compendium of information regarding PDZ complexes demonstrates that dissimilar C-terminal peptides bind to the same PDZ domain, and different PDZ domains can bind the same peptides. A detailed understanding of the PDZ-peptide recognition is needed to elucidate this complexity. To this end, we have designed a family of docking protocols for PDZ domains (termed PDZ-DocScheme) that is based on simulated annealing molecular dynamics and rotamer optimization, and is applicable to the docking of long peptides (20-40 rotatable bonds) to both known PDZ structures and to the more complicated problem of homology models of these domains. The resulting protocol reproduces the structures of PDZ complexes with peptides 4-8 amino acids long within 1-2 A from the experimental structure when the docking is performed to the original structure. If the structure of the target PDZ domain is an apo structure or a homology model, the docking protocol yields structures within 3 A in 9 out of 12 test cases. The automated docking procedure PDZ-DocScheme can serve in the generation of a structural context for validation of PDZ domain specificity from mutagenesis and ligand binding data.

Fast Structure-Based Virtual Ligand Screening Combining FRED, DOCK, and Surflex

A protocol was devised in which FRED, DOCK, and Surflex were combined in a multistep virtual ligand screening (VLS) procedure to screen the pocket of four different proteins. One goal was to evaluate the impact of chaining "freely available packages to academic users" on docking/scoring accuracy and CPU time consumption. A bank of 65 660 compounds including 49 known actives was generated. Our procedure is successful because docking/scoring parameters are tuned according to the nature of the binding pocket and because a shape-based filtering tool is applied prior to flexible docking. The obtained enrichment factors are in line with those reported in recent studies. We suggest that consensus docking/scoring could be valuable to some drug discovery projects. The present protocol could process the entire bank for one receptor in less than a week on one processor, suggesting that VLS experiments could be performed even without large computer resources.

The Limit of Accuracy of Protein Modeling: Influence of Crystal Packing on Protein Structure

The size of the protein database (PDB) makes it now feasible to arrive at statistical conclusions regarding structural effects of crystal packing. These effects are relevant for setting upper practical limits of accuracy on protein modeling. Proteins whose crystals have more than one molecule in the asymmetric unit or whose structures were determined at least twice by X-ray crystallography were paired and their differences analyzed. We demonstrate a clear influence of crystal environment on protein structure, including backbone conformations, hinge-like motions and side-chain conformations. The positions of surface water molecules tend to be variable in different crystal environments while those of ligands are not. Structures determined by independent groups vary more than structures determined by the same authors. The use of different refinement methods is a major source for this effect. Our pair-wise analysis derives a practical limit to the accuracy of protein modeling. For different crystal forms, the limit of accuracy (C(alpha), root-mean-square deviation (RMSD)) is approximately 0.8A for the entire protein, which includes approximately 0.3A due to crystal packing. For organized secondary elements, the upper limit of C(alpha) RMSD is 0.5-0.6A while for loops or protein surface it reaches 1.0A. Twenty percent of exposed side- chains exhibit different chi(1+2) conformations with approximately half of the effect also resulting from crystal packing. A web based tool for analysis and graphic presentation of surface areas of crystal contacts is available (http://ligin.weizmann.ac.il/cryco).

New and fast statistical-thermodynamic method for computation of protein-ligand binding entropy substantially improves docking accuracy

We present a novel method to estimate the contributions of translational and rotational entropy to protein-ligand binding affinity. The method is based on estimates of the configurational integral through the sizes of clusters obtained from multiple docking positions. Cluster sizes are defined as the intervals of variation of center of ligand mass and Euler angles in the cluster. Then we suggest a method to consider the entropy of torsional motions. We validate the suggested methods on a set of 135 PDB protein-ligand complexes by comparing the averaged root-mean square deviations (RMSD) of the top-scored ligand docked positions, accounting and not accounting for entropy contributions, relative to the experimentally determined positions. We demonstrate that the method increases docking accuracy by 10-21% when used in conjunction with the AutoDock docking program, thus reducing the percent of incorrectly docked ligands by 1.4-fold to four-fold, so that in some cases the percent of ligands correctly docked to within an RMSD of 2 A is above 90%. We show that the suggested method to account for entropy of relative motions is identical to the method based on the Monte Carlo integration over intervals of variation of center of ligand mass and Euler angles in the cluster.

Improving Binding Mode Predictions by Docking into Protein-Specifically Adapted Potential Fields

The development of a protein-specifically adapted objective function for docking is described. Structural and energetic information about known protein-ligand complexes is exploited to tailor knowledge-based potentials using a "reverse", protein-based CoMFA-type (=AFMoC) approach. That way, effects due to protein flexibility and information about multiple solvation schemes can be implicitly incorporated. Compared to the application of AFMoC for binding affinity predictions, a Shannon entropy based column filtering of the descriptor matrix and the capping of adapted repulsive potentials within the binding site have turned out to be crucial for the success of this method. The new developed approach (AFMoC(obj)) was validated on a data set of 66 HIV-1 protease inhibitors, for which experimental structural information was available. Convincingly, for ligands with up to 20 rotatable bonds, in more than 75% of all cases a binding mode below 2 A rmsd has been identified on the first scoring rank when AFMoC(obj)-based potentials were used as the objective function in AutoDock. With respect to nonadapted DrugScore or AutoDock fields, the binding mode prediction accuracy was significantly improved by 14%. Noteworthy, very similar results were obtained for training and test set compounds, demonstrating the strength and robustness of this method. Implications of our findings for binding affinity predictions and its usage in virtual screening are further discussed.

Parameter Estimation for Scoring Protein−Ligand Interactions Using Negative Training Data

Surflex-Dock employs an empirically derived scoring function to rank putative protein-ligand interactions by flexible docking of small molecules to proteins of known structure. The scoring function employed by Surflex was developed purely on the basis of positive data, comprising noncovalent protein-ligand complexes with known binding affinities. Consequently, scoring function terms for improper interactions received little weight in parameter estimation, and an ad hoc scheme for avoiding protein-ligand interpenetration was adopted. We present a generalized method for incorporating synthetically generated negative training data, which allows for rigorous estimation of all scoring function parameters. Geometric docking accuracy remained excellent under the new parametrization. In addition, a test of screening utility covering a diverse set of 29 proteins and corresponding ligand sets showed improved performance. Maximal enrichment of true ligands over nonligands exceeded 20-fold in over 80% of cases, with enrichment of greater than 100-fold in over 50% of cases.

Comparing protein-ligand docking programs is difficult

There is currently great interest in comparing protein-ligand docking programs. A review of recent comparisons shows that it is difficult to draw conclusions of general applicability. Statistical hypothesis testing is required to ensure that differences in pose-prediction success rates and enrichment rates are significant. Numerical measures such as root-mean-square deviation need careful interpretation and may profitably be supplemented by interaction-based measures and visual inspection of dockings. Test sets must be of appropriate diversity and of good experimental reliability. The effects of crystal-packing interactions may be important. The method used for generating starting ligand geometries and positions may have an appreciable effect on docking results. For fair comparison, programs must be given search problems of equal complexity (e.g. binding-site regions of the same size) and approximately equal time in which to solve them. Comparisons based on rescoring require local optimization of the ligand in the space of the new objective function. Re-implementations of published scoring functions may give significantly different results from the originals. Ostensibly minor details in methodology may have a profound influence on headline success rates.

Design of Small-Sized Libraries by Combinatorial Assembly of Linkers and Functional Groups to a Given Scaffold: Application to the Structure-Based Optimization of a Phosphodiesterase 4 Inhibitor

Combinatorial chemistry and library design have been reconciled by applying simple medicinal chemistry concepts to virtual library design. The herein reported "Scaffold-Linker-Functional Group" (SLF) approach has the aim to maximize diversity while minimizing the size of a scaffold-focused library with the aid of simple molecular variations in order to identify critical pharmacophoric elements. Straightforward rules define the way of assembling three building blocks: a conserved scaffold, a variable linker, and a variable functional group. By carefully selecting a limited number of functional groups not overlapping in pharmacophoric space, the size of the library is kept to a few hundred. As an application of the SLF approach, a small-sized combinatorial library (320 compounds) was derived from the scaffold of the known phosphodiesterase 4 inhibitor zardaverine. The most interesting analogues were further prioritized for synthesis and enzyme inhibition by FlexX docking to the X-ray structure of the enzyme, leading to a 900-fold increased affinity within nine synthesized compounds and a single screening round.

Comparison of automated docking programs as virtual screening tools

The performance of several commercially available docking programs is compared in the context of virtual screening. Five different protein targets are used, each with several known ligands. The simulated screening deck comprised 1000 molecules from a cleansed version of the MDL drug data report and 49 known ligands. For many of the known ligands, crystal structures of the relevant protein-ligand complexes were available. We attempted to run experiments with each docking method that were as similar as possible. For a given docking method, hit rates were improved versus what would be expected for random selection for most protein targets. However, the ability to prioritize known ligands on the basis of docking poses that resemble known crystal structures is both method- and target-dependent.

ProPose: Steered Virtual Screening by Simultaneous Protein−Ligand Docking and Ligand−Ligand Alignment

The 'model-free' screening engine ProPose implements a general method for performing simultaneous protein-ligand docking, ligand-ligand alignment, pharmacophore queries-and combinations thereof-in order to incorporate a priori information into screening protocols. In this manuscript we describe a case study on herpes simplex virus thymidine kinase, an important antiviral drug target, where we evaluate different approaches for handling a specific type of a priori information, i.e., multiple target structures. We demonstrate that a simultaneous alignment on two target structures--in conjunction with logic operations on interactions and docking constraints derived from protein structure--is an effective means of (i) improving the enrichment of chemical substructures that are compatible with the a priori known ligands, (ii) ensuring the steric fit into the target protein, and (iii) handling target flexibility. The combination of ligand- and receptor-based methods steers the virtual screening by ranking molecules according to the similarity of their interaction pattern with known ligands, thereby--to some extent--outweighing the deficiencies of simple scoring functions often used in initial virtual screening.

Bioinformatics in drug development and assessment

Bioinformatics is playing an increasingly important role in nearly all aspects of drug discovery, drug assessment, and drug development. This growing importance lies not only in the role that bioinformatics plays in handling large volumes of data, but also in the utility of bioinformatics tools to predict, analyze, or help interpret clinical and preclinical findings. This review focuses on describing and evaluating some of the newer or more important bioinformatics resources (i.e., databases and software) that are of growing importance to understanding or predicting drug metabolism, especially with respect to the absorption, distribution, metabolism, excretion, (ADME), and toxicity (T) of both existing drugs and potential drug leads. Detailed descriptions and critical assessments of a number of potentially useful bioinformatics/cheminformatics databases and predictive ADMET software tools are provided. Additionally, several pharmaceutically important applications of both the databases and software are highlighted. Given the rapid growth in this area and the rapid changes that are taking place, a special emphasis is placed on freely available or Web-accessible resources.