Fast 3D molecular superposition and similarity search in databases of flexible molecules

AlphaFold, allosteric, and orthosteric drug discovery: Ways forward

Drug discovery is arguably a highly challenging and significant interdisciplinary aim. The stunning success of the artificial intelligence-powered AlphaFold, whose latest version is buttressed by an innovative machine-learning approach that integrates physical and biological knowledge about protein structures, raised drug discovery hopes that unsurprisingly, have not come to bear. Even though accurate, the models are rigid, including the drug pockets. AlphaFold's mixed performance poses the question of how its power can be harnessed in drug discovery. Here we discuss possible ways of going forward wielding its strengths, while bearing in mind what AlphaFold can and cannot do. For kinases and receptors, an input enriched in active (ON) state models can better AlphaFold's chance of rational drug design success.

Rational Drug Design Rational Drug Design

For the past three decades rationale drug design (RDD) has been developing as an innovative, rapid and successful way to discover new drug candidates. Many strategies have been followed and several targets with diverse structures and different biological roles have been investigated. Despite the variety of computational tools available, one can broadly divide them into two major classes that can be adopted either separately or in combination. The first class involves structure-based drug design, when the target's 3-dimensional structure is available or it can be computationally generated using homology modeling. On the other hand, when only a set of active molecules is available, and the structure of the target is unknown, ligand-based drug design tools are usually used. This review describes some recent advances in rational drug design, summarizes a number of their practical applications, and discusses both the advantages and shortcomings of the various techniques used.

Three-dimensional pharmacophore screening for fentanyl derivatives

Fentanyl is a highly selective μ-opioid receptor agonist with high analgesic activity. Three-dimensional pharmacophore models were built from a set of 50 fentanyl derivatives. These were employed to elucidate ligand-receptor interactions using information derived only from the ligand structure to identify new potential lead compounds. The present studies demonstrated that three hydrophobic regions, one positive ionizable region and two hydrogen bond acceptor region sites located on the molecule seem to be essential for analgesic activity. The results of the comparative molecular field analysis model suggested that both steric and electrostatic interactions play important roles. The contributions from steric and electrostatic fields for the model were 0.621 and 0.379, respectively. The pharmacophore model provides crucial information about how well the common features of a subject molecule overlap with the hypothesis model, which is very valuable for designing and optimizing new active structures.

Rational Drug Design

For the past three decades rationale drug design (RDD) has been developing as an innovative, rapid and successful way to discover new drug candidates. Many strategies have been followed and several targets with diverse structures and different biological roles have been investigated. Despite the variety of computational tools available, one can broadly divide them into two major classes that can be adopted either separately or in combination. The first class involves structure-based drug design, when the target's 3-dimensional structure is available or it can be computationally generated using homology modeling. On the other hand, when only a set of active molecules is available, and the structure of the target is unknown, ligand-based drug design tools are usually used. This review describes some recent advances in rational drug design, summarizes a number of their practical applications, and discusses both the advantages and shortcomings of the various techniques used.

Integrating structure-based and ligand-based approaches for computational drug design

Methods utilized in computer-aided drug design can be classified into two major categories: structure based and ligand based, using information on the structure of the protein or on the biological and physicochemical properties of bound ligands, respectively. In recent years there has been a trend towards integrating these two methods in order to enhance the reliability and efficiency of computer-aided drug-design approaches by combining information from both the ligand and the protein. This trend resulted in a variety of methods that include: pseudoreceptor methods, pharmacophore methods, fingerprint methods and approaches integrating docking with similarity-based methods. In this article, we will describe the concepts behind each method and selected applications.

Novel approach for efficient pharmacophore-based virtual screening: method and applications

Virtual screening is emerging as a productive and cost-effective technology in rational drug design for the identification of novel lead compounds. An important model for virtual screening is the pharmacophore. Pharmacophore is the spatial configuration of essential features that enable a ligand molecule to interact with a specific target receptor. In the absence of a known receptor structure, a pharmacophore can be identified from a set of ligands that have been observed to interact with the target receptor. Here, we present a novel computational method for pharmacophore detection and virtual screening. The pharmacophore detection module is able to (i) align multiple flexible ligands in a deterministic manner without exhaustive enumeration of the conformational space, (ii) detect subsets of input ligands that may bind to different binding sites or have different binding modes, (iii) address cases where the input ligands have different affinities by defining weighted pharmacophores based on the number of ligands that share them, and (iv) automatically select the most appropriate pharmacophore candidates for virtual screening. The algorithm is highly efficient, allowing a fast exploration of the chemical space by virtual screening of huge compound databases. The performance of PharmaGist was successfully evaluated on a commonly used data set of G-Protein Coupled Receptor alpha1A. Additionally, a large-scale evaluation using the DUD (directory of useful decoys) data set was performed. DUD contains 2950 active ligands for 40 different receptors, with 36 decoy compounds for each active ligand. PharmaGist enrichment rates are comparable with other state-of-the-art tools for virtual screening.

Deterministic pharmacophore detection via multiple flexible alignment of drug-like molecules

We present a novel highly efficient method for the detection of a pharmacophore from a set of drug-like ligands that interact with a target receptor. A pharmacophore is a spatial arrangement of physico-chemical features in a ligand that is essential for the interaction with a specific receptor. In the absence of a known three-dimensional (3D) receptor structure, a pharmacophore can be identified from a multiple structural alignment of ligand molecules. The key advantages of the presented algorithm are: (a) its ability to multiply align flexible ligands in a deterministic manner, (b) its ability to focus on subsets of the input ligands, which may share a large common substructure, resulting in the detection of both outlier molecules and alternative binding modes, and (c) its computational efficiency, which allows to detect pharmacophores shared by a large number of molecules on a standard PC. The algorithm was extensively tested on a dataset of almost 80 ligands acting on 12 different receptors. The results, which were achieved using a set of standard default parameters, were consistent with reference pharmacophores that were derived from the bound ligand-receptor complexes. The pharmacophores detected by the algorithm are expected to be a key component in the discovery of new leads by screening large databases of drug-like molecules. A user-friendly web interface is available at http://bioinfo3d.cs.tau.ac.il/pharma. Supplementary material can be found at http://bioinfo3d.cs.tau.ac.il/pharma/reduction/.

Protein-protein interactions: analysis and prediction

Proteins represent the tools and appliances of the cell — they assemble into larger structural elements, catalyze the biochemical reactions of metabolism, transmit signals, move cargo across membrane boundaries and carry out many other tasks. For most of these functions proteins cannot act in isolation but require close cooperation with other proteins to accomplish their task. Often, this collaborative action implies physical interaction of the proteins involved. Accordingly, experimental detection, in silico prediction and computational analysis of protein-protein interactions (PPI) have attracted great attention in the quest for discovering functional links among proteins and deciphering the complex networks of the cell.

Similarity based docking

We have recently introduced GMA, a highly efficient method for flexible molecular alignment. Here we show how this approach can be used to improve docking accuracy and efficiency, in cases where a complex structure of a ligand with the target protein is known. In cases where a known ligand exists, yet the complex structure is unknown it is possible to make use of the advantages offered by this approach, by combining it with standard ligand docking.

TrixX: structure-based molecule indexing for large-scale virtual screening in sublinear time

Structure-based virtual screening today is basically organized as a sequential process where the molecules of a screening library are evaluated for instance with respect to their fit with a biological target. In this paper, we present a novel structure-based screening paradigm avoiding sequential searching and therefore enabling sublinear runtime behavior. We implemented the novel paradigm in the virtual screening tool TrixX and successfully applied it in screening experiments on four targets from relevant therapeutic areas. With the screening paradigm implemented in TrixX, we propose some important extensions and modifications to traditional virtual screening approaches: Instead of processing all compounds in the screening library sequentially, TrixX first analyzes the geometric and physicochemical binding site characteristics and then draws compounds with matching features from a compound catalog. The catalog organizes the compounds by their physicochemical and geometric features making use of relational database technology with indexed tables in order to support efficient queries for compounds with specific features. A key element of the compound catalog is a highly selective geometric descriptor that carries information on the type of functional groups of the compound, their Euclidian distance, the preferred interaction direction of each functional group, and the location of steric bulk around the triangle. In a re-docking experiment with 200 protein-ligand complexes, we could show that TrixX is able to correctly predict the location of ligand functional groups in co-crystallized complexes. In a retrospective virtual screening experiment for four different targets, the enrichment factors of TrixX are comparable to the enrichment factors of FlexX and FlexX-Scan. With computing times clearly below one second per compound, TrixX counts among the fastest virtual screening tools currently available and is nearly two orders of magnitude faster than standard FlexX.

Graph-Based Molecular Alignment (GMA)

We describe a combined 2D/3D approach for the superposition of flexible chemical structures, which is based on recent progress in the efficient identification of common subgraphs and a gradient-based torsion space optimization algorithm. The simplicity of the approach is reflected in its generality and computational efficiency: the suggested approach neither requires precalculated statistics on the conformations of the molecules nor does it make simplifying assumptions on the topology of the molecules being compared. Furthermore, graph-based molecular alignment produces alignments that are consistent with the chemistry of the molecules as well as their general structure, as it depends on both the local connectivities between atoms and the overall topology of the molecules. We validate this approach on benchmark sets taken from the literature and show that it leads to good results compared to computationally and algorithmically more involved methods. The results suggest that, for most practical purposes, graph-based molecular alignment is a viable alternative to molecular field alignment with respect to structural superposition and leads to structures of comparable quality in a fraction of the time.

Geometric accuracy of three-dimensional molecular overlays

This study examines the dependence of molecular alignment accuracy on a variety of factors including the choice of molecular template, alignment method, conformational flexibility, and type of protein target. We used eight test systems for which X-ray data on 145 ligand-protein complexes were available. The use of X-ray structures allowed an unambiguous assignment of bioactive overlays for each compound set. The alignment accuracy depended on multiple factors and ranged from 6% for flexible overlays to 73% for X-ray rigid overlays, when the conformation of the template ligand came from X-ray structures. The dependence of the overlay accuracy on the choice of templates and molecules to be aligned was found to be the most significant factor in six and seven of the eight ligand-protein complex data sets, respectively. While finding little preference for the overlay method, we observed that the introduction of molecule flexibility resulted in a decrease of overlay accuracy in 50% of the cases. We derived rules to maximize the accuracy of alignment, leading to a more than 2-fold improvement in accuracy (from 19% to 48%). The rules also allowed the identification of compounds with a low (<5%) chance to be correctly aligned. Last, the accuracy of the alignment derived without any utilization of X-ray conformers varied from <1% for the human immunodeficiency virus data set to 53% for the trypsin data set. We found that the accuracy was directly proportional to the product of the overlay accuracy from the templates in their bioactive conformations and the chance of obtaining the correct bioactive conformation of the templates. This study generates a much needed benchmark for the expectations of molecular alignment accuracy and shows appropriate usages and best practices to maximize hypothesis generation success.

Combining 4D Pharmacophore Generation and Multidimensional QSAR: Modeling Ligand Binding to the Bradykinin B2 Receptor

We recently reported the development of two receptor-modeling concepts (software Quasar and Raptor) based on multidimensional quantitative structure-activity relationships (QSAR) and allowing for the explicit simulation of induced fit. As the identification of the bioactive configuration of ligand molecules in such studies is all but unambiguous, each compound may be represented by an ensemble of different conformations, orientations, stereoisomers, and protonation states, leading to a 4D data set. In this account, we present a novel technology (software Symposar) allowed to automatically generate a 4D pharmacophore as input for multidimensional QSAR. Symposar aligns ligands utilizing fuzzylike 2D-subfeature mapping and, subsequently, a Monte Carlo search on a 3D similarity grid. The two-step concept (4D pharmacophore generation and quantification of ligand binding by multidimensional QSAR) was applied to 186 compounds binding to the bradykinin B2 receptor. The prediction of their binding affinity by means of the Quasar and Raptor technologies allowed for consensus scoring and generated topologically and quantitatively consistent receptor models. These converged at a cross-validated r2 of 0.752 and 0.815 and yielded a predictive r2 of 0.784 and 0.853 for a test set (for Quasar and Raptor, respectively).

BRUTUS: Optimization of a grid-based similarity function for rigid-body molecular superposition. II. Description and characterization

Finding novel lead molecules is one of the primary goals in early phases of drug discovery projects. However, structurally dissimilar compounds may exhibit similar biological activity, and finding new and structurally diverse lead compounds is difficult for computer algorithms. Molecular energy fields are appropriate for finding structurally novel molecules, but they are demanding to calculate and this limits their usefulness in virtual screening of large chemical databases. In our approach, energy fields are computed only once per superposition and a simple interpolation scheme is devised to allow coarse energy field lattices having fewer grid points to be used without any significant loss of accuracy. The resulting processing speed of about 0.25 s per conformation on a 2.4 GHz Intel Pentium processor allows the method to be used for virtual screening on commonly available desktop machines. Moreover, the results indicate that grid-based superposition methods could be efficiently used for the virtual screening of compound libraries.

Representation of target-bound drugs by computed conformers: implications for conformational libraries

BACKGROUND

The increasing number of known protein structures provides valuable information about pharmaceutical targets. Drug binding sites are identifiable and suitable lead compounds can be proposed. The flexibility of ligands is a critical point for the selection of potential drugs. Since computed 3D structures of millions of compounds are available, the knowledge of their binding conformations would be a great benefit for the development of efficient screening methods.

RESULTS

Integration of two public databases allowed superposition of conformers for 193 approved drugs with 5507 crystallised target-bound counterparts. The generation of 9600 drug conformers using an atomic force field was carried out to obtain an optimal coverage of the conformational space. Bioactive conformations are best described by a conformational ensemble: half of all drugs exhibit multiple active states, distributed over the entire range of the reachable energy and conformational space.A number of up to 100 conformers per drug enabled us to reproduce the bound states within a similarity threshold of 1.0 angstroms in 70% of all cases. This fraction rises to about 90% for smaller or average sized drugs.

CONCLUSION

Single drugs adopt multiple bioactive conformations if they interact with different target proteins. Due to the structural diversity of binding sites they adopt conformations that are distributed over a broad conformational space and wide energy range. Since the majority of drugs is well represented by a predefined low number of conformers (up to 100) this procedure is a valuable method to compare compounds by three-dimensional features or for fast similarity searches starting with pharmacophores. The underlying 9600 generated drug conformers are downloadable from the Super Drug Web site 1. All superpositions are visualised at the same source. Additional conformers (110,000) of 2400 classified WHO-drugs are also available.

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.

Influence of conformation on the representation of small flexible molecules at low resolution: alignment of endothiapepsin ligands

In this contribution, we discuss a molecular representation mode for the generation of reduced descriptions of flexible molecules in various conformations. The representations of the endothiapepsin ligands are constituted by graphs of peaks obtained through a hierarchical merging algorithm which combines the location of promolecular electron density (ED) maxima with the decomposition of the molecular structures into fragments. The representations are then aligned through the use of a Monte Carlo/Simulated Annealing procedure. The evaluation function of the alignment solutions is based on the local density values of the peaks and their inter-distances. The applications show that the alignment of a given molecule onto itself, in a different conformation, is successful when a pseudo-topological path length is considered in the evaluation function of the solution, while Cartesian distances are more adapted to the scoring for alignments of two different molecules in their co-crystallized conformation. Results are compared with the available literature data.

FLAME: A Program to Flexibly Align Molecules

Herein, we describe a method to flexibly align molecules (FLAME = FLexibly Align MolEcules). FLAME aligns two molecules by first finding maximum common pharmacophores between them using a genetic algorithm. The resulting alignments are then subjected to simultaneous optimizations of their internal energies and an alignment score. The utility of the method in pairwise alignment, multiple molecule flexible alignment, and database searching was examined. For pairwise alignment, two carboxypeptidase ligands (Protein Data Bank codes and ), two estrogen receptor ligands ( and ), and two thrombin ligands ( and ) were used as test sets. Alignments generated by FLAME starting from CONCORD structures compared very well to the X-ray structures (average root-mean-square deviation = 0.36 A) even without further minimization in the presence of the protein. For multiple flexible alignments, five structurally diverse D3 receptor ligands were used as a test set. The FLAME alignment automatically identified three common pharmacophores: a base, a hydrogen-bond acceptor, and a hydrophobe/aromatic ring. The best alignment was then used to search the MDDR database. The search results were compared to the results using atom pair and Daylight fingerprint similarity. A similar database search comparison was also performed using estrogen receptor modulators. In both cases, hits identified by FLAME were structurally more diverse compared to those from the atom pair and Daylight fingerprint methods.

Comparing the Performance of FLUFF-BALL to SEAL-CoMFA with a Large Diverse Estrogen Data Set: From Relevant Superpositions to Solid Predictions

In this work a template-based molecular mechanistic superposition algorithm FLUFF (Flexible Ligand Unified Force Field) and an accompanying local coordinate QSAR method BALL (Boundless Adaptive Localized Ligand) are validated against the benchmark techniques SEAL (Steric and Electrostatic Alignment) and CoMFA (Comparative Molecular Field Analysis) using a large diverse set of 245 xenoestrogens extracted from the EDKB (Endocrine Disruptor Knowledge Base) maintained by NCTR (National Centre for Toxicological Research). The results indicate that FLUFF is capable of generating relevant superpositions not only for BALL but also for CoMFA, as both techniques give predictive QSAR models. When the BALL and CoMFA methods are compared, it is clear that the BALL algorithm met or even exceeded the results of the standard 3D-QSAR method CoMFA using alignments either from the tailor-made superposition technique FLUFF or the reference method SEAL. The FLUFF-BALL method can be easily automated, and it is computationally light, providing thus a good computational "sieve" capable of fast screening of large molecule libraries.

BRUTUS: Optimization of a Grid-Based Similarity Function for Rigid-Body Molecular Superposition. 1. Alignment and Virtual Screening Applications

We have developed a fast grid-based algorithm, BRUTUS, for rigid-body molecular superposition and similarity searching. BRUTUS aligns molecules using field information derived from charge distributions and van der Waals shapes of the compounds. Molecules can have similar biological properties if their charge distributions and shapes are similar, even though they have different chemical structures; that is, BRUTUS can identify compounds possessing similar properties, regardless of their structures. In this paper, we present two applications of BRUTUS. First, BRUTUS was used to superimpose five sets of inhibitors. Second, two sets of known inhibitors were searched from a database, and the results were analyzed using self-organizing maps. We demonstrate that BRUTUS is successful in superimposing compounds using molecular fields and, importantly, is fast and accurate enough for virtual screening of chemical databases using a standard personal computer. This fast and efficient molecular-field-based algorithm is applicable for virtual screening of structurally diverse, active molecules.

Conformation Mining: An Algorithm for Finding Biologically Relevant Conformations

Discovering essential features shared by active compounds, an important step in drug-design, is complicated by conformational flexibility. We present a new algorithm to efficiently mine the conformational space of multiple actives and find small subsets of conformations likely to be biologically relevant. The approach identifies chemical and steric similarities between actives, providing insight into features important for binding when structural data are absent. Validation studies (thrombin and CDK2 data) produce alignments similar to protein-based alignments.

Comparison of 2D Similarity and 3D Superposition. Application to Searching a Conformational Drug Database

In a database of about 2000 approved drugs, represented by 10(5) structural conformers, we have performed 2D comparisons (Tanimoto coefficients) and 3D superpositions. For one class of drugs the correlation between structural resemblance and similar action was analyzed in detail. In general Tanimoto coefficients and 3D scores give similar results, but we find that 2D similarity measures neglect important structural/funtional features. Examples for both over- and underestimation of similarity by 2D metrics are discussed. The required additional effort for 3D superpositions is assessed by implementation of a fast algorithm with a processing time below 0.01 s and a more sophisticated approach (0.5 s per superposition). According to the improvement of similarity detection compared to 2D screening and the pleasant rapidity on a desktop PC, full-atom 3D superposition will be an upcoming method of choice for library prioritization or similarity screening approaches.

Ligand-based structural hypotheses for virtual screening

The majority of drug targets for small molecule therapeutics are proteins whose three-dimensional structure is not known to sufficient resolution to permit structure-based design. All three-dimensional QSAR approaches have a requirement for some hypothesis of ligand conformation and alignment, and predictions of molecular activity critically depend on this ligand-based binding site hypothesis. The molecular similarity function used in the Surflex docking system, coupled with quantitative pressure to minimize overall molecular volume, forms an effective objective function for generating hypotheses of bioactive conformations of sets of small molecules binding to their cognate proteins. Results are presented, assessing utility of the method for ligands of the serotonin, histamine, muscarinic, and GABA(A) receptors. The Surflex similarity module (Surflex-Sim) was able, in each case, to distinguish true ligands from random compounds using models constructed from just two or three known ligands. True positive rates of 60% were achieved with false positive rates of 0-3%; the theoretical enrichment rates were over 150-fold compared with random screening. The methods are practically applicable for rational design of ligands and for high-throughput virtual screening and offer competitive performance to many structure-based docking algorithms.

Novel technologies for virtual screening

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