Virtual screening—an overview

Prediction of 'drug-likeness'

Recent developments in combinatorial chemistry and high-throughput screening have dramatically increased the scale on which drug discovery programs are carried out. Along with these advances has come a need for automated methods of determining which compounds from a library should be synthesized and screened. These methods range from simple counting schemes to sophisticated machine learning techniques such as neural networks. While many of these methods have performed well in validation studies, the field is still in its formative stage. This paper reviews a number of computational techniques for identifying drug-like molecules and examines challenges facing the field.

Chemical database techniques in drug discovery

Chemical databases are becoming a powerful tool in drug discovery. Database searches based on possible requirements for biological activity can identify compounds that might be suitable for further analysis or indicate novel ways to achieve the desired activity. What considerations are involved in the construction and searching of chemical databases?

Virtual Screening in Lead Discovery: A Viewpoint

Virtual screening (VS) methods have emerged as an adaptive response to massive throughput synthesis and screening technologies. Based on the structure-permeability paradigm, the Lipinski rule of five has become a standard property filtering protocol for VS. Three possible VS scenarios with respect to optimising binding affinity and pharmacokinetic properties are discussed. The parsimony principle for selecting candidate leads for further optimisation is advocated.

Computation of the physio-chemical properties and data mining of large molecular collections

Very large data sets of molecules screened against a broad range of targets have become available due to the advent of combinatorial chemistry. This information has led to the realization that ADME (absorption, distribution, metabolism, and excretion) and toxicity issues are important to consider prior to library synthesis. Furthermore, these large data sets provide a unique and important source of information regarding what types of molecular shapes may interact with specific receptor or target classes. Thus, the requirement for rapid and accurate data mining tools became paramount. To address these issues Pharmacopeia, Inc. formed a computational research group, The Center for Informatics and Drug Discovery (CIDD).* In this review we cover the work done by this group to address both in silico ADME modeling and data mining issues faced by Pharmacopeia because of the availability of a large and diverse collection (over 6 million discrete compounds) of drug-like molecules. In particular, in the data mining arena we discuss rapid docking tools and how we employ them, and we describe a novel data mining tool based on a ID representation of a molecule followed by a molecular sequence alignment step. For the ADME area we discuss the development and application of absorption, blood-brain barrier (BBB) and solubility models. Finally, we summarize the impact the tools and approaches might have on the drug discovery process.

Development of a virtual screening method for identification of "frequent hitters" in compound libraries

A computer-based method was developed for rapid and automatic identification of potential "frequent hitters". These compounds show up as hits in many different biological assays covering a wide range of targets. A scoring scheme was elaborated from substructure analysis, multivariate linear and nonlinear statistical methods applied to several sets of one and two-dimensional molecular descriptors. The final model is based on a three-layered neural network, yielding a predictive Matthews correlation coefficient of 0.81. This system was able to correctly classify 90% of the test set molecules in a 10-times cross-validation study. The method was applied to database filtering, yielding between 8% (compilation of trade drugs) and 35% (Available Chemicals Directory) potential frequent hitters. This filter will be a valuable tool for the prioritization of compounds from large databases, for compound purchase and biological testing, and for building new virtual libraries.

Virtual screening and fast automated docking methods

Recent advances in high-throughput protein structure determination and in computational chemistry have refocused attention on virtual screening and fast automated docking methods. This review provides a brief introduction to the basic ideas and outlines computational tools currently used. We also provide several examples of where virtual screening has proved successful, highlighting the usefulness of the approach.

High-throughput crystallography for lead discovery in drug design

Knowledge of the three-dimensional structures of protein targets now emerging from genomic data has the potential to accelerate drug discovery greatly. X-ray crystallography is the most widely used technique for protein structure determination, but technical challenges and time constraints have traditionally limited its use primarily to lead optimization. Here, we describe how significant advances in process automation and informatics have aided the development of high-throughput X-ray crystallography, and discuss the use of this technique for structure-based lead discovery.

Evaluation of docking functions for protein-ligand docking

Docking functions are believed to be the essential component of docking algorithms. Both physically and statistically based functions have been proposed, but there is no consensus about their relative performances. Here, we propose an evaluation approach based on exhaustive enumeration of all possible docking solutions obtained with a discretized description of a rigid docking process. We apply the approach to study both molecular mechanics and statistical potentials. It is found that the statistical potential evaluated is less effective than the AMBER molecular mechanics function to provide an accurate description of the docking process when the exact experimental coordinates are used. However, when coordinates of crystal structures obtained with analogous ligands are used, similar performances are obtained in both cases. Possible reasons for the successes and failures of both docking schemes have been uncovered using linear discriminant analysis, on the basis of a set of physicochemical descriptors capturing the main physical effects at play during protein-ligand docking. In both types of potentials steric effects appear critical to obtain a successful docking. Our results also indicate that neglecting desolvation effects and the explicit treatment of hydrogen bonds are the main source of the failures observed with the molecular mechanics potential. On the other hand, detailed consideration of steric interactions, with a careful treatment of dispersive forces, seems to be needed when using statistical potentials derived from a structural database. The possibility of filtering combinatorial libraries in order to maximize the probability of correct docking is discussed.

4D-QSAR analysis of a set of ecdysteroids and a comparison to CoMFA modeling

The ecdysteroid-responsive Drosophila melanogaster B(II) cell line is a prototypical homologous inducible gene expression system. A training set of 71 ecdysteroids, for which the -log(EC(50)) potencies in the ecdysteroid-responsive B(II) cell line were measured, was used to construct 4D-QSAR models. Four nearly equivalent optimum 4D-QSAR models, for two modestly different alignments, were identified (Q(2) = 0.76-0.80). These four models, together with two CoMFA models, were used in consensus modeling to arrive at a three-dimensional pharmacophore. The C-2 and C-22 hydroxyls are identified as hydrogen-bond acceptor sites which enhance activity. A hydrophobic site near C-12 is consistent with increasing activity. The side-chain substituents at C-17 are predicted to adopt semiextended "active" conformations which could fit into a cylinder-shaped binding pocket lined largely with nonpolar residues for enhanced activity. A test set of 20 ecdysteroids was used to evaluate the QSAR models. Two 4D-QSAR models for one alignment were identified to be superior to the others based on having the smallest average residuals of prediction for the prediction set (0.69 and 1.13 -log[EC(50)] units). The correlation coefficients of the optimum 4D-QSAR models (R(2) = 0.87 and 0.88) are nearly the same as those of the best CoMFA model (R(2) = 0.92) determined for the same training set. However, the cross-validation correlation coefficient of the CoMFA model is less significant (Q(2) = 0.59) than those of the 4D-QSAR models (Q(2) = 0.80 and 0.80).

The impact of informatics and computational chemistry on synthesis and screening

High-throughput synthesis and screening technologies have enhanced the impact of computational chemistry on the drug discovery process. From the design of targeted, drug-like libraries to 'virtual' optimization of potency, selectivity and ADME/Tox properties, computational chemists are able to efficiently manage costly resources and dramatically shorten drug discovery cycle times. This review will describe some of the successful strategies and applications of state-of-the-art algorithms to enhance drug discovery, as well as key points in the drug discovery process where computational methods can have, and have had, greatest impact.

Ligand-protein database: linking protein-ligand complex structures to binding data

In computational structure-based drug design, the scoring functions are the cornerstones to the success of design/discovery. Many approaches have been explored to improve their reliability and accuracy, leading to three families of scoring functions: force-field-based, knowledge-based, and empirical. The last family is the most widely used in association with docking algorithms because of its speed, even though such empirical scoring functions produce far too many false positives to be fully reliable. In this work, we describe a World Wide Web accessible database that gathers the structural information from known complexes of the PDB with experimental binding data. This database, the Ligand-Protein DataBase (LPDB), is designed to allow the selection of complexes based on various properties of receptors and ligands for the design and parametrization of new scoring functions or to assess and improve existing ones. Moreover, for each complex, a continuum of ligand positions ranging from the crystallographic position to points on the surface of the protein receptor allows an assessment of the energetic behavior of particular scoring functions.

Are free energy calculations useful in practice? A comparison with rapid scoring functions for the p38 MAP kinase protein system

Precise thermodynamic integration free energy simulations have been applied to a congeneric series of 16 inhibitors to the p38 MAP kinase protein for which the experimental binding data (IC(50)) is known. The relative free energy of binding for each compound has been determined. For comparison, the same series of compounds have also scored using the best rapid scoring functions used in database screening. From the results of these calculations, we find (1) that precise free energy simulations allow predictions that are reliable and in good agreement with experiment; (2) that predictions of lower reliability, but still in good qualitative agreement with experiment, can be obtained using the OWFEG free energy grid method, at a much lower computational cost; (3) and that other methods, not based on free energy simulations yield results in much poorer agreement with experiment. A new predictive index, which measures the reliability of a prediction method in the context of normal use, is defined and calculated for each scoring method. Predictive indices of 0.84, 0.56, 0.04, -0.05, and 0.25 are calculated for thermodynamic integration, OWFEG, ChemScore, PLPScore, and Dock Energy Score, respectively, where +1.0 is perfect correct prediction, -1.0 is perfect incorrect prediction, and 0.0 is random.

High-throughput structural proteomics using x-rays

Following the recent sequencing of the human genome, the focus has shifted from the DNA level to the protein level, with the goal of elucidating function. Technical developments in x-ray crystallography mean that the crystal structures of these new proteins can now been determined at an unprecedented rate, which assists in functional analysis and rational drug-design programmes.

How does consensus scoring work for virtual library screening? An idealized computer experiment

It has been reported recently that consensus scoring, which combines multiple scoring functions in binding affinity estimation, leads to higher hit-rates in virtual library screening studies. This method seems quite independent to the target receptor, the docking program, or even the scoring functions under investigation. Here we present an idealized computer experiment to explore how consensus scoring works. A hypothetical set of 5000 compounds is used to represent a chemical library under screening. The binding affinities of all its member compounds are assigned by mimicking a real situation. Based on the assumption that the error of a scoring function is a random number in a normal distribution, the predicted binding affinities were generated by adding such a random number to the "observed" binding affinities. The relationship between the hit-rates and the number of scoring functions employed in scoring was then investigated. The performance of several typical ranking strategies for a consensus scoring procedure was also explored. Our results demonstrate that consensus scoring outperforms any single scoring for a simple statistical reason: the mean value of repeated samplings tends to be closer to the true value. Our results also suggest that a moderate number of scoring functions, three or four, are sufficient for the purpose of consensus scoring. As for the ranking strategy, both the rank-by-number and the rank-by-rank strategy work more effectively than the rank-by-vote strategy.

Virtual screening of combinatorial libraries across a gene family: in search of inhibitors of Giardia lamblia guanine phosphoribosyltransferase

Parasitic protozoa lack the ability to synthesize purine nucleotides de novo, relying instead on purine salvage enzymes for their survival. Guanine phosphoribosyltransferase (GPRT) from the protozoan parasite Giardia lamblia is a potential target for rational antiparasitic drug design, based on the experimental evidence, which indicates the lack of interconversion between adenine and guanine nucleotide pools. The present study is a continuation of our efforts to use three-dimensional structures of parasitic phosphoribosyltransferases (PRTs) to design novel antiparasitic agents. Two micromolar phthalimide-based GPRT inhibitors were identified by screening the in-house phthalimide library. A combination of structure-based scaffold selection using virtual library screening across the PRT gene family and solid phase library synthesis led to identification of smaller (molecular weight, <300) ligands with moderate to low specificity for GPRT; the best inhibitors, GP3 and GP5, had K(i) values in the 23 to 25 microM range. These results represent significant progress toward the goal of designing potent inhibitors of purine salvage in Giardia parasites. As a second step in this process, altering the phthalimide moiety to optimize interactions in the guanine-binding pocket of GPRT is expected to lead to compounds with promising activity against G. lamblia PRT.

High-throughput docking for lead generation

Recent improvements in flexible docking technology may lead to a bigger role for computational methods in lead discovery. Although fast and accurate computational prediction of binding affinities for an arbitrary molecule is still beyond the limits of current methods, the docking and screening procedures can select small sets of likely lead candidates from large libraries of either commercially or synthetically available compounds.

High-throughput screening approaches for investigating drug metabolism and pharmacokinetics

1. High-throughput screening approaches have been adopted throughout the pharmaceutical industry to aid in the rapid discovery of new chemical entities. Because it is now well recognized that the selection of a robust candidate requires a balance of potency, safety and pharmacokinetics, the role of drug metabolism departments has widened from their traditional one of supporting drug development to include the screening of compounds during the discovery process. To put drug metabolism and pharmacokinetic (DMPK) studies in context, the evolving role of DMPK screening in the drug discovery strategy of pharmaceutical companies will be discussed and a generalized approach will be presented. 2. With the increasing numbers of compounds requiring screening, DMPK optimization methods have had to be adapted for high throughput. There have been many developments in this field over the past decade and this review will focus on the high-throughput DMPK screening methodologies used today and in the recent past. 3. In vitro and in silico (computer-based) methods have proven most amenable to high-throughput approaches and these will firm the bulk of the review, but some advances with in vivo methods will also be discussed. As there has been a vast increase in published material on the topic of high-throughput DMPK methodologies in the past 10 years, it would be impossible to cover every method in detail, so this review will concentrate on the key areas and refer the reader to other, more detailed reviews wherever possible. 4. Most high-throughput methods would not be possible without the enabling technologies of computing, automation, new sample preparation technologies, and highly sensitive and selective detection systems, and these will also be reviewed. 5. The advantages and disadvantages of the screening methods will be presented, in particular the issue of handling the false-positives and -negatives that arise. 6. In concluding the review, future developments in this field will be discussed along with key issues that will need to be addressed.

Simple selection criteria for drug-like chemical matter

A simple pharmacophore point filter has been developed that discriminates between drug-like and nondrug-like chemical matter. It is based on the observation that nondrugs are often underfunctionalized. Therefore, a minimum count of well-defined pharmacophore points is required to pass the filter. The application of the filter results in 66-69% of subsets of the MDDR database to be classified as drug-like. Furthermore, 61-68% of subsets of the CMC database are classified as drug-like. In contrast, only 36% of the ACD are found to be drug-like. While these results are not quite as good as those obtained with recently described neural net approaches, the method used here has clear advantages. In contrast to a neural net approach and also in contrast to decision tree methods described recently, the pharmacophore filter has been developed by using "chemical wisdom" that is unbiased from fitting the structural content of specific drug databases to prediction models. Similar to decision tree methods, the pharmacophore point filter provides a detailed structural reason for the classification of each molecule as drug or nondrug. The pharmacophore point filter results are compared to neural net filter results. A statistically significant overlap between compounds recognized as drug-like validates both approaches. The pharmacophore point filter complements neural net approaches as well as property profiling approaches used as drug-likeness filters in compound library analysis and design.

Miniaturized HTS technologies - uHTS

The transition from slow, manual, low-throughput screening to industrialized robotic ultra-high throughput screening (uHTS) in the past few years has made it possible to screen hundreds of thousands of chemical entities against a biological target in a short time-frame. The need to minimize the cost of screening has been addressed primarily by reducing the volume of sample to be screened. This, in turn, has resulted in the miniaturization of HTS technology as a whole. Miniaturization requires new technologies and strategies for compound handling, assay development, assay adaptation, liquid handling and automation in addition to refinement of the technologies used for detection systems and data management. This review summarizes current trends in the field of uHTS and illustrates the technological developments that are necessary to enable the routine application of miniaturized uHTS systems within an industrial environment.

A fast algorithm for searching for molecules containing a pharmacophore in very large virtual combinatorial libraries

We present a new algorithm for identifying molecules that display a pharmacophore, or in general a structural motif, by efficiently constructing and screening huge virtual combinatorial libraries of diverse compounds. The uniqueness of this algorithm is its ability to build and screen libraries of ca. 10(18) 3D molecular conformations within a reasonable time scale, thereby increasing the chemical space that can be virtually screened by many orders of magnitude. The algorithm may be used to design new molecules that display a desired pharmacophore on predefined sets of chemical scaffolds. This is demonstrated herein by screening a library of backbone cyclic peptides to find candidate peptido- and proteinomimetics.

Fingerprint scaling increases the probability of identifying molecules with similar activity in virtual screening calculations

Results of systematic virtual screening calculations using a structural key-type fingerprint are reported for compounds belonging to 14 activity classes added to randomly selected synthetic molecules. For each class, a fingerprint profile was calculated to monitor the relative occupancy of fingerprint bit positions. Consensus bit patterns were determined consisting of all bits that were always set on in compounds belonging to a specific activity class. In virtual screening calculations, scale factors were applied to each consensus bit position in fingerprints of query molecules. This technique, called "fingerprint scaling", effectively increases the weight of consensus bit positions in fingerprint comparisons. Although overall prediction accuracy was satisfactory using unscaled calculations, scaling significantly increased the number of correct predictions but only slightly increased the rate of false positives. These observations suggest that fingerprint scaling is an attractive approach to increase the probability of identifying molecules with similar activity by virtual screening. It requires the availability of a series of related compounds and can be easily applied to any keyed fingerprint representation that associates bit positions with specific molecular features.

Customized versus universal scoring functions: application to class I MHC-peptide binding free energy predictions

A tailor-made free energy scoring method (Fresno) has been compared to six universal scoring functions (Chemscore, Dock, FlexX, Gold, Pmf, Score) for predicting the binding affinity of 26 peptides to the class I human major histocompatibility protein HLA-B*2705. Fresno clearly outperforms all six universal scoring functions.

Improved scoring of ligand-protein interactions using OWFEG free energy grids

A new approach to rapidly score protein-ligand interactions is tested on several protein-ligand systems. Results using this approach - the OWFEG free energy grid - are quite promising and are generally in better agreement with experiment (in some cases much better) than those obtained employing scoring techniques currently in wide use. The OWFEG free energy grid is generated from a one-window free energy perturbation MD simulation (Pearlman, D. A. J. Med. Chem. 1999, 42, 4313-4324). The OWFEG approach is applied to three protein systems: IMPDH, MAP kinase p38, and HIV-1 aspartyl protease. OWFEG scores are compared to experimental K(i) and IC50 data in each case. Empirical scoring functions applied to the same systems for comparison include ChemScore, Piecewise Linear Potential (PLP), and Dock energy score.

Design, docking, and evaluation of multiple libraries against multiple targets

We present a general approach to the design, docking, and virtual screening of multiple combinatorial libraries against a family of proteins. The method consists of three main stages: docking the scaffold, selecting the best substituents at each site of diversity, and comparing the resultant molecules within and between the libraries. The core "divide-and-conquer" algorithm for side-chain selection, developed from an earlier version (Sun et al., J Comp Aided Mol Design 1998;12:597-604), provides a way to explore large lists of substituents with linear rather than combinatorial time dependence. We have applied our method to three combinatorial libraries and three serine proteases: trypsin, chymotrypsin, and elastase. We show that the scaffold docking procedure, in conjunction with a novel vector-based orientation filter, reproduces crystallographic binding modes. In addition, the free-energy-based scoring procedure (Zou et al., J Am Chem Soc 1999;121:8033-8043) is able to reproduce experimental binding data for P1 mutants of macromolecular protease inhibitors. Finally, we show that our method discriminates between a peptide library and virtual libraries built on benzodiazepine and tetrahydroisoquinolinone scaffolds. Implications of the docking results for library design are explored.

Library design for NMR-based screening

Recent advances in NMR-based screening methods have made it possible to screen larger libraries of molecules with higher throughput. However, experience shows that intelligent library design is important if NMR screening is to succeed in aiding our discovery of potent and useful lead compounds. This review presents the current state-of-the-art methodologies for designing primary and follow-up libraries for NMR screening. Diversity, drug-likeness and combinatorial libraries are discussed, and the inherent pitfalls of the NMR approach are addressed.

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

Three different database docking programs (Dock, FlexX, Gold) have been used in combination with seven scoring functions (Chemscore, Dock, FlexX, Fresno, Gold, Pmf, Score) to assess the accuracy of virtual screening methods against two protein targets (thymidine kinase, estrogen receptor) of known three-dimensional structure. For both targets, it was generally possible to discriminate about 7 out of 10 true hits from a random database of 990 ligands. The use of consensus lists common to two or three scoring functions clearly enhances hit rates among the top 5% scorers from 10% (single scoring) to 25-40% (double scoring) and up to 65-70% (triple scoring). However, in all tested cases, no clear relationships could be found between docking and ranking accuracies. Moreover, predicting the absolute binding free energy of true hits was not possible whatever docking accuracy was achieved and scoring function used. As the best docking/consensus scoring combination varies with the selected target and the physicochemistry of target-ligand interactions, we propose a two-step protocol for screening large databases: (i) screening of a reduced dataset containing a few known ligands for deriving the optimal docking/consensus scoring scheme, (ii) applying the latter parameters to the screening of the entire database.

High-throughput and virtual screening: core lead discovery technologies move towards integration

In addition to high-throughput screening (HTS), the main lead discovery technology employed by most pharmaceutical companies today is virtual screening (VS). Although the two techniques have somewhat different philosophical origins, they contain many synergies that can potentially enhance the lead discovery process. Here, we describe many of the latest developments in VS technology with particular emphasis on their potential impact on HTS in, for example, focussed screening and data mining. In addition, we highlight key issues that need to be addressed before the potential of such efforts can be fully realized.

ADME/PK as part of a rational approach to drug discovery

Rational drug discovery requires an early appraisal of all factors impacting on the likely success of a drug candidate in the subsequent preclinical, clinical and commercial phases of drug development. The study of absorption, distribution, metabolism, excretion and pharmacokinetics (ADME/PK) has developed into a relatively mature discipline in drug discovery through the application of well-established in vitro and in vivo methodologies. The availability of improved analytical and automation technologies has dramatically increased our ability to dissect out the fundamentals of ADME/PK through the development of increasingly powerful in silico methods. This is fuelling a shift away from the traditional, empirical nature of ADME/PK towards a more rational, in cerebro approach to drug design.

NMR techniques for characterization of ligand binding: utility for lead generation and optimization in drug discovery

Over the last ten years, nmr spectroscopy has evolved into an important discipline in drug discovery. Initially, nmr was most useful as a technique to provide structural information regarding protein drug targets and target-ligand interactions. More recently, it has been shown that nmr may be used as an alternative method for identification of small molecule ligands that bind to protein drug targets. High throughput implementation of these experiments to screen small molecule libraries may lead to identification of potent and novel lead compounds. In this review, we will use examples from our own research to illustrate how nmr experiments to characterize ligand binding may be used to both screen for novel compounds during the process of lead generation, as well as provide structural information useful for lead optimization during the latter stages of a discovery program.

Designing targeted libraries with genetic algorithms

In combinatorial synthesis, molecules are assembled by linking chemically similar fragments. Because the number of available chemical fragments often greatly exceeds the number that can be used in one synthetic experiment, one needs a rational method for choosing a subset of desirable fragments. If a combinatorial library is to be targeted against a particular biological activity, virtual screening methods can be used to predict which molecules in a virtual library are most likely to be active. When the number of possible molecules in a virtual library is very large, genetic algorithms (GAs) or simulated annealing can be used to quickly find high-scoring molecules by sampling a small subset of the total combinatorial space. We previously demonstrated how a GA can be used to select a subset of fragments for a combinatorial library, and we used topology-based methods of scoring. Here we extend that earlier work in three ways. (1) We demonstrate use of the GA with 3D scoring methods developed in our laboratory. (2) We show that the approach of assembling libraries from fragments in high-scoring molecules is a reasonable one. (3) We compare results from a library-based GA to those from a molecule-based GA.

Diversity screening versus focussed screening in drug discovery

Which strategy is best for hit identification? Making the right choices in the capital-intensive world of modern drug discovery can make the difference between success and expensive failure. Keeping an open mind to all the options is essential. Two well-established strategies are diversity-based and focussed screening. This review will provide contrasting viewpoints highlighting the strengths and deficiencies of each approach, as well as some insights into why both strategies are likely to have a place in the research armoury of a successful drug company.

Developing a dynamic pharmacophore model for HIV-1 integrase

We present the first receptor-based pharmacophore model for HIV-1 integrase. The development of "dynamic" pharmacophore models is a new method that accounts for the inherent flexibility of the active site and aims to reduce the entropic penalties associated with binding a ligand. Furthermore, this new drug discovery method overcomes the limitation of an incomplete crystal structure of the target protein. A molecular dynamics (MD) simulation describes the flexibility of the uncomplexed protein. Many conformational models of the protein are saved from the MD simulations and used in a series of multi-unit search for interacting conformers (MUSIC) simulations. MUSIC is a multiple-copy minimization method, available in the BOSS program; it is used to determine binding regions for probe molecules containing functional groups that complement the active site. All protein conformations from the MD are overlaid, and conserved binding regions for the probe molecules are identified. Those conserved binding regions define the dynamic pharmacophore model. Here, the dynamic model is compared to known inhibitors of the integrase as well as a three-point, ligand-based pharmacophore model from the literature. Also, a "static" pharmacophore model was determined in the standard fashion, using a single crystal structure. Inhibitors thought to bind in the active site of HIV-1 integrase fit the dynamic model but not the static model. Finally, we have identified a set of compounds from the Available Chemicals Directory that fit the dynamic pharmacophore model, and experimental testing of the compounds has confirmed several new inhibitors.

A novel approach for the virtual screening and rational design of anticancer compounds

A topological substructural approach to molecular design (TOSS-MODE) has been introduced for the selection and design of anticancer compounds. A quantitative model that discriminates anticancer compounds from the inactive ones in a training series was obtained. This model permits the correct classification of 91.43% of compounds in an external prediction set with only 1.43% of false actives and 7. 14% of false inactives. The model developed is then used in a simulation of a virtual search for Ras FTase inhibitors; 87% of the Ras FTase inhibitors used in this simulated search were correctly classified, thus indicating the ability of the TOSS-MODE model of finding lead compounds with novel structures and mechanism of action. Finally, a series of carbonucleosides was designed, and the compounds were classified as active/inactive anticancer compounds by using the model developed here. From the compounds so-designed, 20 were synthesized and evaluated experimentally for their antitumor effects on the proliferation of murine leukemia cells (L1210/0) and human T-lymphocyte cells (Molt4/C8 and CEM/0); 80% of these compounds were well-classified, as active or inactive, and only two pairs of isomeric compounds were false actives. The chloropurine derivatives were the most active compounds, especially compounds 6c, d.

Combinatorial library design: maximizing model-fitting compounds within matrix synthesis constraints

The use of combinatorial chemistry has become commonplace within the pharmaceutical industry. Less widespread but gaining in popularity is the derivation of activity models from the high-throughput assays of these libraries. Such models are then used as filters during the design of refined daughter libraries. The design of these second generation libraries, which efficiently test and conform to the derived activity model from the large space of virtual possibilities, remains an area of considerable research. We present here a computationally efficient method for the design of optimally dense (in model matching compounds) synthetic matrices from in silico virtual libraries.

New approach to molecular docking and its application to virtual screening of chemical databases

This paper describes the validation of a molecular docking method and its application to virtual database screening. The code flexibly docks ligand molecules into rigid receptor structures using a tabu search methodology driven by an empirically derived function for estimating the binding affinity of a protein-ligand complex. The docking method has been tested on 70 ligand-receptor complexes for which the experimental binding affinity and binding geometry are known. The lowest energy geometry produced by the docking protocol is within 2.0 A root mean square of the experimental binding mode for 79% of the complexes. The method has been applied to the problem of virtual database screening to identify known ligands for thrombin, factor Xa, and the estrogen receptor. A database of 10,000 randomly chosen "druglike" molecules has been docked into the three receptor structures. In each case known receptor ligands were included in the study. The results showed good separation between the predicted binding affinities of the known ligand set and the database subset.

Diversity measures for enhancing ADME admissibility of combinatorial libraries

For general screening libraries, structural diversity descriptors and drug-likeness indicators still do not guarantee the in vivo bioavailability for the candidates, which is considered a major bottleneck in drug development. Early prediction of pharmacokinetics (log P, log D), metabolism, and toxicity makes it possible to deal with ADME (adsorption, distribution, metabolism, excretion) related diversity as an extension to the classical diversity concepts. It opens several new possibilities for optimization of a discovery library before doing any experimental screening. This new diversity concept is demonstrated on a subset of MeDiverse, which is a diverse collection of pharmacologically relevant compounds selected from our in-house library. From consideration of the ADME interface in living systems, virtual secondary libraries of metabolites and retrometabolites (prodrugs) can be generated. These additional libraries readily enhance both the structural and ADME related diversity. This new opportunity in library design can substantially improve the success rate for in vivo lead generation from in vitro hits.

Bit-string methods for selective compound acquisition

Selective compound acquisition programs need to ensure that the compounds that are chosen do not contain undesirable functionality. This is easy to achieve if a supplier is prepared to provide unambiguous structure representations for the compounds that they have available: this paper discusses selection techniques that can be used when a supplier is prepared to make available only fragment bit-string representations for the compounds in their catalog. Experiments with three databases and three types of bit-string show that a simple k-nearest-neighbor searching method provides a surprisingly effective, although far from perfect, way of selecting compounds when only bit-string representations are available. A second approach, based on the use of a fragment weighting scheme analogous to those used in substructural analysis studies, proved to be noticeably less effective in operation.