key: cord-0965641-4a7pk5bd
authors: Berry, Michael; Fielding, Burtram; Gamieldien, Junaid
title: Chapter 27 Practical Considerations in Virtual Screening and Molecular Docking
date: 2015-12-31
journal: Emerging Trends in Computational Biology, Bioinformatics, and Systems Biology
DOI: 10.1016/b978-0-12-802508-6.00027-2
sha: 7fa7dbba5f5ad4414ef4bd62aefa2683879e0dd6
doc_id: 965641
cord_uid: 4a7pk5bd

Abstract Molecular docking has become an important common component of the drug discovery toolbox, and its relative low-cost implications and perceived simplicity of use has stimulated an everincreasing popularity within academic communities. The inherent “garbage-in-garbage-out” defect of molecular docking, however, leads a lot of researchers to dedicate countless hours to the identification of hit compounds that later prove to be inactive. Several considerations that can greatly improve the success and enrichment of true bioactive hit compounds are commonly overlooked at the initial stages of a molecular docking study. This chapter will cover several of these considerations, including protonation states, active site waters, separating actives from decoys, consensus docking and molecular mechanics generalized-Born/surface area (MM-GBSA) rescoring, and incorporation of pharmacophoric constraints, in an attempt to clarify what is, in fact, very complicated and inherent difficulties of a structure-based drug design study.

are, therefore, also capable of filtering through, and ranking, large databases of compounds in virtual screening, where the highest-ranked binding energies should correspond to a potential lead (Phatak et al., 2009) .

Used on its own, molecular docking is, however, plagued with weaknesses. The static nature of the receptor is a primary fault where the dynamic nature of the biological structures is not considered. Limitations in sampling algorithms and imperfections in scoring functions also lead to the generation of both false positives and false negatives (Lill, 2011; Wang et al., 2003; Brooijmans and Kuntz, 2003; Alvarez, 2004) and the requirement for training sets in various algorithms often leads to accuracy being highly target dependent (Warren et al., 2006) . These inherent flaws are further exacerbated by user oversights and errors. This chapter will detail several practical aspects to consider prior to commencing a virtual screening study, while simultaneously providing a theoretical explanation of docking and scoring. This review will provide guidelines, but there is no "one-rule-fits-all" in molecular docking. Most docking programs have varying methods to deal with each topic discussed and describing details of each program is outside the scope of this review. It must also be taken into account that every receptor is different and the ability to replicate experimental and physiological findings is highly system dependent.

Although most receptor preparation tools accurately complete processes that were not undertaken during X-ray crystal structure refinement, it is important to understand these processes and make adjustments where necessary. The most common receptor preparation procedures include adding hydrogens and atom-type charges, but it is also important to ensure that missing side-chains are added, missing bonds and molecule chain breaks are detected and fixed, bond orders are assigned, and where alternate locations are present, the atoms with highest frequencies must be selected. Other, more complicated, procedures in receptor preparation include accurate prediction of protonation states and identifying which water molecules (if any) should remain in the receptor structure. All of these procedures maximize the biological realism in the modeled system, which leads to the identification of a higher proportion of true bioactives.

The resolution of most crystal structures does not provide information on the location of hydrogens, commonly referred to as the protonation state (ten Brink and Exner, 2009) . The accurate prediction of the correct protonation state, especially within the binding interface, is crucial to accurately predict the correct binding mode and, to a greater extent, binding affinity (Kalliokoski et al., 2009; Fornabaio et al., 2003; Onufriev and Alexov, 2013) . This incorrect prediction of binding mode and affinity will inevitably lead to the identification of false positives, while true bioactives are missed (Onufriev and Alexov, 2013) . It is notable to point out that force field-based scoring functions are more susceptible to incorrect protonation states in comparison to knowledge-based scoring functions (Onufriev and Alexov, 2013) . Assigning the incorrect protonation states further alters the state of hydrogen bond donors and acceptors, which substantially limits the accurate prediction of protein-ligand interactions (Polgár and Keserü, 2005) .

Side-chains of ionizable amino acids can further vary their protonation states within a receptor depending on the local environment and pH. Ligand binding can also be accompanied by proton gain or release (Petukh et al., 2013) but this is almost never incorporated into a molecular docking study (Onufriev and Alexov, 2013) . One study pointed out that a residue's protonation state cannot be accurately replicated, as protons are not static and are readily transferred between molecules (Fornabaio et al., 2003) . The quantum mechanical simulations necessary to replicate proton movements are far beyond the scope and capabilities of molecular docking and at best, the protonation state, or an ensemble of protonation states, that is most suitable to ligand binding must be identified.

Histidine (His) provides a unique problem in terms of residue protonation, as it can be protonated in three different conformations. The imidazole ring of the His side-chain can be protonated in a neutral confirmation at the ε-nitrogen or the δ-nitrogen or in a charged (+1) conformation where both the εand δ-nitrogens are protonated (Kim et al., 2013) . To further complicate the correct conformation of the imidazole side-chain ring, ambiguities in crystal structures often switch the carbon and nitrogen, creating an additional three rotameric conformations, termed "flipped" (Glusker et al., 1994) . His also represents a weaker base, and for this reason, determining the protonation state is more complicated than for other ionizable residues and must be determined individually (Waszkowycz et al., 2011) . In the case of His, analysis of hydrogen bonding networks is likely to yield the most detail about the correct side-chain protonation.

The dynamic nature of a receptor means the protonation states of ionizable residues are constantly changing. In order to accurately predict the conformation of a ligand binding to a receptor, the protonation state of the receptor must be relevant to the bound conformation and in correspondence with crystal data (i.e., absence of steric clashes and hydrogen bonds occurring at expected locations) and in accordance with the pH of the experimental conditions. Assigning protonation states to Asp, Glu, Arg, and Lys during receptor preparation is generally straightforward, with deprotonated acids (Asp and Glu) and protonated bases (Arg and Lys) (Kim et al., 2013; Waszkowycz et al., 2011) . This is, however, a generalization and not a rule, and the microenvironment of the residue and physiological pH of the receptor must be taken into careful consideration. Calculating the theoretical pK a of these residues at the physiological pH is possibly the most straightforward mechanism to determine or estimate their protonation state (Polgár and Keserü, 2005) .

As scoring functions are highly dependent on the correct receptor protonation state, it can be assumed that a scoring function will favor the correct protonation state by scoring it above the incorrect state (Onufriev and Alexov, 2013) . This provides a mechanism to accurately predict the correct protonation state within an ensemble of pregenerated receptor states. The correct replication of hydrogen bond positions between ligand and receptor, as seen in the crystal structure or detailed in the literature, will further suggest the accurate placement of residue protons (Krieger et al., 2012; Hooft et al., 1996) . Observable steric clashes between a ligand and receptor, after protonation, will further suggest incorrect proton placement (Word et al., 1999; Krieger et al., 2012) . This approach will only account for ionizable groups within the binding interface and will not be able to account for the entire receptor, but this remains a far more attractive strategy than ignoring the issue entirely.

In summary, in order to accurately approximate a receptor's protonation state, the identification of its physiological pH is key. Second, calculated pK a values for ionizable residues enables determination of the protonation state according to the given pK a at the specified pH. Third, crystal structures and known, experimentally identified bioactives can yield a wealth of knowledge on the protonation state of a receptor by scoring function analysis and inspection of steric clashes and hydrogen bonding networks between ligand and receptor. Given these guidelines, the techniques used to accurately predict the correct protonation state of a receptor are largely dependent on the class of receptor being studied. For this reason, the techniques applied must be accurately verified for the receptor under investigation before virtual library screening.

Active site water molecules are key determinants in ligand-receptor binding (Thilagavathi and Mancera, 2010; Barillari et al., 2007) . Not only can they mediate hydrogen bonding between ligand and receptor, but their contribution to entropic and enthalpic changes are significant (Lie et al., 2011; Cheng et al., 2012; Kroemer, 2007) . In a virtual screening context, the addition of water (an explicit solvent) is frequently neglected, as the intensive computational simulations required does not permit the rapid screening required for large libraries, often seen in high-throughput virtual screens, and accounting for water molecules in docking remains a significant challenge (Cheng et al., 2012; Huang and Shoichet, 2008; Schneider and Fechner, 2005) .

The position of water molecules within an active site are also highly variable (Santos et al., 2009) , and to account for them as static in nature would be biased toward ligands that complement the specific orientation and prejudice those that would physiologically replace the water molecules, leading to a drastic increase in false negatives (Kroemer, 2007) . Several reports claim to more accurately predict the binding mode of crystal structure inhibitors by incorporating water molecules within the active site (Lemmon and Meiler, 2013) . While these studies do possess a high degree of merit, the inclusion of waters within the active site greatly decreases the volume of the pocket and thereby the possible conformations that the ligand may assume, which is further biased toward the correct conformation (Lie et al., 2011; Hartshorn et al., 2007) . As there is a constant compromise between speed and accuracy in a high-throughput virtual screen, however, the presence of active site waters can greatly increase ligand enrichment. It is, therefore, important to determine which waters, if any, must be kept during a virtual screen and exclude those that are nonessential.

An initial step to assess the importance of active site waters would be to attempt to replicate the binding mode of experimental structures in the absence of explicit waters. If the accuracy is diminished by the absence of waters in the binding site, it is important to select which waters are pivotal to binding. Waters that are not hydrogen bonded to the receptor, and those that are located outside the binding pocket (more than 5Å ), will obviously have little effect on ligand binding and can therefore be removed (Huang and Shoichet, 2008) . Waters that possess three hydrogen bonds with the receptor, or those with low B-factors, are likely to be highly stable within the pocket and should be included in docking studies, as these waters may prove difficult to displace by ligand binding and likely function to stabilize the protein binding site (Yang et al., 2006; Hornak et al., 2006) . Waters that form hydrogen bond bridges between the ligand and receptor are also likely to be important in ligand binding. This may, however, be highly ligand-specific and its importance in virtual screening, where a diverse set of ligand classes are under study, must be properly assessed and validated. Where essential water molecules are included in a virtual screen they should, ideally, be treated as flexible (Huang and Shoichet, 2008) . It is also important to bear in mind that the accuracy of a docking algorithm may be highly dependent on the parameterization of the algorithm and suitability toward the class of receptor and inhibitor, which will be discussed later in this chapter.

It is commonly accepted that there is no "first-in-class" algorithm or molecular docking software for the prediction of correct ligand-binding pose or relative free energy of binding. Molecular docking algorithms are often calibrated on a training set of experimental ligand-protein complexes and accuracy of these docking programs is often highly dependent on the training set used (Ballester and Mitchell, 2010) . This highlights the importance of confirming that the docking software used for virtual screening is capable of replicating the binding mode of known, experimental inhibitors for the class of receptor studied (Lim et al., 2011; Kroemer, 2007) . To improve ligand enrichment in a virtual screening context, the docking algorithm selected must be properly validated for the class of receptor under investigation. Of course, in a virtual screen, where hundreds of thousands to millions of compounds are potentially being screened, validating for each class of potential inhibitor would be impossible, but accurate validation must be undertaken with the largest obtainable data set of true experimental leads, where the binding pose is known. A root-mean-square deviation (RMSD) below 2Å for heavy atoms (excluding hydrogens) between the experimental structure and predicted pose of docking is a well-defined benchmark to assess the accuracy of molecular docking sampling algorithms (Houston and Walkinshaw, 2013) .

A highly useful benchmarking strategy and metric to gauge the success of a molecular docking program is the ability to differentiate true actives from decoys. The Database of Useful Decoys-Enhanced (DUD-E; http://dude.docking.org/gener ate) can generate decoys for an active compound (Mysinger et al., 2012) . DUD generates decoys based on cheminformatic properties, including molecular weight, logP, number of rotatable bonds, and number of hydrogen bond donors and acceptors. As these decoys are not intended to bind to the target receptor, they are topologically distinct from the active inhibitors, thereby serving as suitable negative controls. The enrichment of the docking program can be assessed by its ability to rank true actives above decoy ligands (Mysinger et al., 2012) .

Virtual screening simulations are typically performed on static structures, and it has previously been demonstrated that the use of a holo (ligand-bound) conformation provides better enrichment when compared to apo or homology modeled receptors (McGovern and Shoichet, 2003) . Given this, addressing protein flexibility can substantially improve enrichment but remains one of the most challenging aspects of molecular docking. There are currently two approaches to incorporate the dynamic nature of protein structures; flexible receptor docking and ensemble docking (Lill, 2013) . These approaches have shown to improve enrichment in docking studies (Craig et al., 2010) , but the compromise between speed and accuracy must be heavily weighted in high-throughput virtual screens.

Flexible docking most often only incorporates side-chains of residues within the active site and therefore does not cover the dynamic range of protein conformations (Meng et al., 2011) . It has been demonstrated that only a small number of side-chains within a binding pocket undergo structural changes upon ligand binding. This study suggested that, within 85% of studied receptors, only three or fewer side-chains exhibited movements upon ligand binding and further developed a scale of sidechain flexibility (Lys > Arg, Gln, Met > Glu, Ile, Leu > Asn, Thr, Val, Tyr, Ser, His, Asp > Cys, Trp, Phe; Najmanovich et al., 2000) . Utilizing this scale, it may be possible to identify which side-chains within a pocket must be made flexible and which may be left static, although the ability to accurately enrich active ligands must be displayed.

In ensemble docking, the ensemble of rigid structures can be generated by a molecular dynamic simulation where snapshots are isolated from the trajectory or when several structures are available from crystallography or nuclear magnetic resonance (NMR) experimental studies. There are two distinct classes of ensemble docking. In the first method, several protein conformations are generated prior to a docking screen and each ligand is docked into each receptor independently (Carlson, 2002; Carlson and McCammon, 2000; Barril and Morley, 2005) , thereby introducing receptor flexibility by multiple docking runs (Henzler and Rarey, 2010) . This is, of course, computationally inefficient and the time required to conduct a screen increases with every protein structure included in the ensemble. The conformational diversity is also limited to the conformational representations included in the ensemble (B-Rao et al., 2009 ). The second method assesses an ensemble of protein structures in a single docking screen (B-Rao et al., 2009 ). This method either unites ensemble structures or uses a receptor grid averaged over all protein structures, and therefore reduces computational cost considerably (Totrov and Abagyan, 2008; Knegtel et al., 1997; Henzler and Rarey, 2010) . To identify a suitable ensemble of structures to incorporate in a docking run, an enrichment docking screen of known actives can be performed. Both ensemble and flexible receptor docking is described in greater detail in several reviews (Cavasotto and Abagyan, 2004; Carlson, 2002; Therrien et al., 2014; Henzler and Rarey, 2010) .

Molecular dynamic (MD) simulations is considered to be the most accurate method to determine the stability of a ligand within a binding pocket, while accounting for full side-chain and backbone flexibility and incorporating solvent effects (Marco and Gago, 2007; Alonso et al., 2006) . Several docking studies have utilized MD simulations to confirm results obtained from docking studies. However, the intense computational costs make it practical for only a small set of ligands ( € Osterberg and Å qvist, 2005; Han, 2012; Mukherjee et al., 2011; Segura-Cabrera et al., 2013) .

Scoring functions have been highlighted as the major weakness of molecular docking (Yang et al., 2005; Warren et al., 2006; Wang et al., 2003) . As these functions are solely responsible for selecting and ranking the correct ligand pose within the binding site from the many possible conformations generated by the sampling algorithm, it can potentially lead to identification of an incorrect pose. The integration of a consensus approach to sampling and scoring, incorporating several algorithms to each task, has shown to greatly improve ligand enrichment in virtual screening and identifying the correct pose of experimental structures (Teramoto and Fukunishi, 2007; Houston and Walkinshaw, 2013; Kukol, 2011; Yang et al., 2005; Charifson et al., 1999; Plewczynski et al., 2011) . Consensus scoring compensates for deficiencies in individual scoring functions and thereby improves the overall performance (Teramoto and Fukunishi, 2007) , with the inclusion of a single extra scoring function being sufficient to improve binding affinity predictions (Chang et al., 2010) . A similar technique to consensus scoring is the approach of consensus sampling, which is less well characterized. A recent study by Houston and Walkinshaw, 2013 utilized three sampling algorithms from Dock (Ewing et al., 2001) , Autodock (Morris et al., 2009) , and Autodock Vina (Trott and Olson, 2010) to identify the experimental pose of a diverse set of ligands. The study achieved an accuracy of 82%, compared to the 55%-64% accuracy of using a single algorithm (Houston and Walkinshaw, 2013) . In this study, a consensus result was confirmed when independently predicted poses were within an RMSD cutoff of 2Å , the same distance defined as correct sampling in comparison to experimental structures (Houston and Walkinshaw, 2013) .

The approach of employing several algorithms to identify the correct pose with subsequent consensus scoring to identify top-ranked ligands can greatly improve the enrichment rate in a virtual screening context. The major cost of this approach is the increase in false negatives, which are therefore missed and do not progress to experimental testing. In an academic setting, or a lab where resources are limited, this is an acceptable consequence, as the quality of the results is more vital in a virtual screening context. The improvement in the identified hit list, with a decrease in false positives and subsequent decrease in resource waste, would largely compensate for the increase in false negatives (Houston and Walkinshaw, 2013) .

Various elements of binding free energy, including long-range electrostatics, desolvation upon binding, and entropic contributions, are poorly defined in conventional scoring functions utilized in molecular docking (Rastelli et al., 2010a) . These terms are better defined by more rigorous and computationally intensive calculations included in techniques such as free energy perturbation (FEP; Kollman, 1993) , thermodynamic integration (TI; Lybrand et al., 1986) , linear response (LR; Å qvist et al., 1994), molecular mechanics Poisson-Boltzamann/surface area (MM-PBSA; Kuhn and Kollman, 2000) and molecular mechanics generalized-Born/surface area (MM-GBSA; Kollman et al., 2000) . Of these, MM-PBSA and MM-GBSA are faster by several orders of magnitude, making them favorable techniques for the rescoring and reranking of hit lists identified by virtual screening. As these techniques are computationally efficient and yield high correlations with experimental binding energies, the general opinion that docking results should be further analyzed by more advanced approaches is increasing (Rastelli et al., 2010a , Sgobba et al., 2012 .

MM-PBSA and MM-GBSA previously required an ensemble of snapshots, generated by an MD simulation of the protein-ligand complex in water. This has been replaced by the use of a continuum implicit solvent model with a single minimized protein-ligand structure. This technique has given excellent correlations with experimental data (Guimarães and Cardozo, 2008; Greenidge et al., 2013) and is comparable with the more time-consuming and computer-intensive approach of averaging MD simulations in water (Rastelli et al., 2010a) . The use of a single energyminimized structure with a continuum implicit solvent has further improved the enrichment of virtual screens and can successfully discriminate between true binders and decoys (Rastelli et al., 2010a) . Explicit solvent models have further shown to decrease this correlation (Greenidge et al., 2013) .

MM-PBSA and MM-GBSA are force field-based methods that use a combination of molecular mechanics (MM) energies, polar and nonpolar solvation terms, and an entropy term to calculate the free energy of binding (ΔG bind ; Massova and Kollman, 2000; Kollman et al., 2000) from the change between the bound complex (ΔG com ) and unbound receptor (ΔG rec ) and ligand (ΔG lig ) in solution [Eq. (27.1); Rastelli et al., 2010a; Guimarães and Cardozo, 2008] : Greenidge et al., 2013) and a Poisson-Boltzamn (PB) distribution in MM-PBSA , where the nonpolar contribution is usually calculated as a linear function of the solvent accessible surface area (Hou et al., 2011a; Greenidge et al., 2013) . With these functions, the binding free energy (ΔG bind ) is calculated: In most studies, the entropy (T. ΔS) term is neglected, as its calculation can be a major source of error (Rastelli et al., 2010b) and does not always improve the prediction accuracy (Hou et al., 2011a; Guimaraes, 2012) ; however, some researchers do still advocate its use (e.g., Lafont et al., 2007) . When comparing the PB and GB methods in calculation of solvation terms, the PB model is theoretically more rigorous and computationally intensive than GB but does not always give a stronger correlation with experimental binding free energy. The GB model is also more efficient and faster at ranking binding affinities of ligands, making it more suitable in a virtual screening context (Hou et al., 2011a (Hou et al., , 2011b Huang et al., 2010; Li et al., 2010) . MM-GBSA has further been shown to be a more attractive option than the computationally heavy FEP and TI methodologies, as it can be as accurate and computationally more efficient, and handle structurally more diverse ligands because it requires no training set (Guimarães and Cardozo, 2008) . In conclusion, MM-GBSA provides excellent correlation with experimental binding energy, improved enrichment in virtual screening of compound databases, is computationally suitable for medium-throughput screening or reranking a defined hit list and provides more accurate docking poses (Hou et al., 2011b) . With this, the rescoring of docking complexes using MM-GBSA has emerged as a computationally important approach in structure-based drug design (Guimaraes, 2012) .

A pharmacophore is defined as an ensemble of structural features that are necessary for molecular recognition (Guner, 2000) . These features predominantly include hydrophobic moieties and hydrogen bond donors and acceptors, but may also include aromatic rings, cations, and anions. A pharmacophore model can be used prior to a docking study to reduce the size of a ligand library, or it can be used to filter hits following a virtual screen. These pharmacophoric features can be defined by an ensemble of known, active inhibitors where features that are frequently repeated are included in the pharmacophore model (Yang, 2010) , or they can be defined by the natural substrate. Identifying ligands that are able to replicate the interactions made between the natural substrate and a receptor can greatly improve the success and enrichment of a virtual screen. An example of substrate derived pharmacophoric constraints is the three chymotrypsin-like protease (3CL pro ) of coronaviruses. The S 1 pocket in this family of proteases has an absolute specificity for glutamine, which is mediated by a hydrogen bond between the substrate and His163, deep in the pocket. The S 2 pocket forms a deep hydrophobic region that displays preference for a hydrophobic moiety and the Glu166 residue increases substrate specificity via an additional hydrogen bond (Zheng et al., 2007; Chuck et al., 2011; Shoichet, 2004; Schapira et al., 2003) . These pharmacophoric features have been extensively used to identify novel inhibitors of the 3CL pro (Jacobs et al., 2013) .

Despite its limitations, molecular docking has yielded the discovery of novel leads (Shoichet and Kobilka, 2012; Wang and Ekins, 2006) and, if used correctly, the speed and cost effectiveness at which molecular docking screens can be conducted can provide an excellent starting point in a project with few to no compelling leads (Alvarez, 2004) . Possibly the most important consideration to make when commencing a structure-based drug design study is a question of project design, especially if the user is a beginner in the field. The more prior knowledge and availability of published data, the greater the chance of success in the project where proper scrutiny of available literature is essential. The availability of high-resolution crystals or NMR structures of the receptor are paramount prior to a virtual screen, as homology models have been proven to yield low enrichment when compared to holo or apo experimental structures. Holo structures have further proven to improve enrichment, and the state of the experimental structure should therefore be taken into account (McGovern and Shoichet, 2003) . Efficient characterization of the active or allosteric binding site is essential. Detailed understanding of the location and flexibility of side-chains within the pocket, the presence or absence of active site waters, and protonation states of ionisable residues will contribute greatly to enrichment in a virtual screen. The availability of known actives will also allow for essential benchmarking, validation, and potential generation of an effective pharmacophore model. It is important to characterize what class of inhibitors these actives belong to. Molecular docking is not capable of replicating covalent interactions between ligand and receptor, and therefore covalent inhibitors should be excluded. Large peptidomimetics are also difficult to dock with conventional docking methodologies. This is directly related to the inaccuracy of docking algorithms to predict the correct conformation of compounds with increased number of rotatable bonds. With each rotatable bond, the conformational space that must be sampled increases dramatically and thereby reduces the chance of successfully predicting the correct pose. Ligands in a molecular docking screen, therefore, should be limited to eight rotatable bonds (Houston and Walkinshaw, 2013) . A final consideration covered in this chapter is the use of consensus scoring and sampling. This has been shown to greatly improve enrichment with MM-GBSA rescoring, yielding high correlations with experimental evidence and should be considered in a virtual screening context (Teramoto and Fukunishi, 2007; Chang et al., 2010; Houston and Walkinshaw, 2013; Hou et al., 2011b) .

Combining docking and molecular dynamic simulations in drug design

High-throughput docking as a source of novel drug leads

The Process of Structure-Based Drug Design

A new method for predicting binding affinity in computer-aided drug design

A machine learning approach to predicting protein-ligand binding affinity with applications to molecular docking

Classification of water molecules in protein binding sites

Unveiling the full potential of flexible receptor docking using multiple crystallographic structures

Managing protein flexibility in docking and its applications

Molecular recognition and docking algorithms

Protein flexibility and drug design: how to hit a moving target

Accommodating protein flexibility in computational drug design

Protein flexibility in ligand docking and virtual screening to protein kinases

Virtual screening for HIV protease inhibitors: a comparison of AutoDock 4 and Vina

Consensus scoring: a method for obtaining improved hit rates from docking databases of three-dimensional structures into proteins

Structure-based virtual screening for drug discovery: a problem-centric review

Profiling of substrate specificities of 3C-like proteases from group 1, 2a, 2b, and 3 coronaviruses

Ensemble docking into multiple crystallographically derived protein structures: an evaluation based on the statistical analysis of enrichments

DOCK 4.0: search strategies for automated molecular docking of flexible molecule databases

Simple, intuitive calculations of free energy of binding for protein-ligand complexes. 2. Computational titration and pH effects in molecular models of neuraminidase-inhibitor complexes

Crystal Structure Analysis for Chemists and Biologists

MM/GBSA binding energy prediction on the PDBbind data set: successes, failures, and directions for further improvement

MM-GB/SA Rescoring of Docking Poses. Computational Drug Discovery and Design

MM-GB/SA rescoring of docking poses in structurebased lead optimization

Pharmacophore Perception, Development, and Use in Drug Design. International University Line

Docking and molecular dynamics simulations of celastrol binding to p23

Diverse, high-quality test set for the validation of protein-ligand docking performance

Pursuit of Fully Flexible Protein-Ligand Docking: Modeling the Bilateral Mechanism of Binding

Positioning hydrogen atoms by optimizing hydrogen-bond networks in protein structures

HIV-1 protease flaps spontaneously close to the correct structure in simulations following manual placement of an inhibitor into the open state

Assessing the performance of the molecular mechanics/Poisson Boltzmann surface area and molecular mechanics/generalized Born surface area methods. II. The accuracy of ranking poses generated from docking

Assessing the performance of the MM/PBSA and MM/GBSA methods. 1. The accuracy of binding free energy calculations based on molecular dynamics simulations

Consensus docking: improving the reliability of docking in a virtual screening context

Exploiting ordered waters in molecular docking

Scoring functions and their evaluation methods for protein-ligand docking: recent advances and future directions

Discovery, synthesis, and structure-based optimization of a series of N-(tert-butyl)-2-(N-arylamido)-2-(pyridin-3-yl) acetamides (ML188) as potent noncovalent small molecule inhibitors of the severe acute respiratory syndrome coronavirus (SARS-CoV) 3CL protease

Lead optimization via high-throughput molecular docking

The effect of ligand-based tautomer and protomer prediction on structure-based virtual screening

Effects of histidine protonation and rotameric states on virtual screening of M. tuberculosis RmlC

Molecular docking to ensembles of protein structures

Free energy calculations: applications to chemical and biochemical phenomena

Calculating structures and free energies of complex molecules: combining molecular mechanics and continuum models

Assignment of Protonation States in Proteins and Ligands: Combining Pka Prediction with Hydrogen Bonding Network Optimization. Computational Drug Discovery and Design

Structure-based drug design: docking and scoring

Binding of a diverse set of ligands to avidin and streptavidin: an accurate quantitative prediction of their relative affinities by a combination of molecular mechanics and continuum solvent models

Consensus virtual screening approaches to predict protein ligands

A geometric approach to macromolecule-ligand interactions

Compensating enthalpic and entropic changes hinder binding affinity optimization

Towards ligand docking including explicit interface water molecules

Test MM-PB/SA on true conformational ensembles of protein À ligand complexes

Molecular docking with ligand attached water molecules

Efficient incorporation of protein flexibility and dynamics into molecular docking simulations

Virtual screening in drug design

Structure-based and ligand-based virtual screening of novel methyltransferase inhibitors of the dengue virus

Theoretical Calculation of Relative Binding Affinity in Host-Guest Systems

Overcoming the inadequacies or limitations of experimental structures as drug targets by using computational modeling tools and molecular dynamics simulations

Combined molecular mechanical and continuum solvent approach (MM-PBSA/GBSA) to predict ligand binding

Information decay in molecular docking screens against holo, apo, and modeled conformations of enzymes

Molecular docking: a powerful approach for structure-based drug discovery

AutoDock4 and AutoDockTools4: automated docking with selective receptor flexibility

Inhibitors of SARS-3CLpro: virtual screening, biological evaluation, and molecular dynamics simulation studies

Directory of useful decoys, enhanced (DUD-E): better ligands and decoys for better benchmarking

Side-chain flexibility in proteins upon ligand binding

Protonation and pK changes in protein-ligand binding

Exploring blocker binding to a homology model of the open hERG K < sup > +</sup > channel using docking and molecular dynamics methods

The role of protonation states in ligand-receptor recognition and binding

High-Throughput and in Silicon Screenings in Drug Discovery

VoteDock: consensus docking method for prediction of protein-ligand interactions

Virtual screening for β-secretase (BACE1) inhibitors reveals the importance of protonation states at Asp32 and Asp228

Fast and accurate predictions of binding free energies using MM-PBSA and MM-GBSA

Fast and accurate predictions of binding free energies using MM-PBSA and MM-GBSA

State-of-the-art in ligand-based virtual screening

Role of water in molecular docking simulations of cytochrome P450 2D6

Discovery of Diverse Thyroid Hormone Receptor Antagonists by High-Throughput Docking

Computer-based de novo design of drug-like molecules

Repurposing of FDA-approved drugs for the discovery of inhibitors of dengue virus NS2B-NS3 protease by docking, consensus scoring, and molecular dynamics simulations

Application of a post-docking procedure based on MM-PBSA and MM-GBSA on single and multiple protein conformations

Virtual screening of chemical libraries

Structure-based drug screening for G-protein-coupled receptors

Influence of protonation, tautomeric, and stereoisomeric states on protein À ligand docking results

Supervised consensus scoring for docking and virtual screening

Docking Ligands into Flexible and Solvated Macromolecules. 7. Impact of Protein Flexibility and Water Molecules on Docking-based Virtual Screening Accuracy

Ligand À Protein cross-docking with water molecules

Flexible ligand docking to multiple receptor conformations: a practical alternative

AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading

Computer Applications in Pharmaceutical Research and Development

Comparative evaluation of 11 scoring functions for molecular docking

A critical assessment of docking programs and scoring functions

Outstanding challenges in protein-ligand docking and structure-based virtual screening

Asparagine and glutamine: using hydrogen atom contacts in the choice of side-chain amide orientation

Pharmacophore modeling and applications in drug discovery: challenges and recent advances

Consensus scoring criteria for improving enrichment in virtual screening

Drug Design targeting the main protease, the Achilles heel of Coronaviruses

Insight into the activity of SARS main protease: molecular dynamics study of dimeric and monomeric form of enzyme

Hit identification and optimization in virtual screening: practical recommendations based on a critical literature analysis