ParaMol: A Package for Automatic Parameterization of Molecular Mechanics Force Fields.

The ensemble of structures generated by molecular mechanics (MM) simulations is determined by the functional form of the force field employed and its parameterization. For a given functional form, the quality of the parameterization is crucial and will determine how accurately we can compute observable properties from simulations. While accurate force field parameterizations are available for biomolecules, such as proteins or DNA, the parameterization of new molecules, such as drug candidates, is particularly challenging as these may involve functional groups and interactions for which accurate parameters may not be available. Here, in an effort to address this problem, we present ParaMol, a Python package that has a special focus on the parameterization of bonded and nonbonded terms of druglike molecules by fitting to ab initio data. We demonstrate the software by deriving bonded terms' parameters of three widely known drug molecules, viz. aspirin, caffeine, and a norfloxacin analogue, for which we show that, within the constraints of the functional form, the methodologies implemented in ParaMol are able to derive near-ideal parameters. Additionally, we illustrate the best practices to follow when employing specific parameterization routes. We also determine the sensitivity of different fitting data sets, such as relaxed dihedral scans and configurational ensembles, to the parameterization procedure, and discuss the features of the various weighting methods available to weight configurations. Owing to ParaMol's capabilities, we propose that this software can be introduced as a routine step in the protocol normally employed to parameterize druglike molecules for MM simulations.

Detection of secondary binding sites in proteins using fragment screening.

Proteins need to be tightly regulated as they control biological processes in most normal cellular functions. The precise mechanisms of regulation are rarely completely understood but can involve binding of endogenous ligands and/or partner proteins at specific locations on a protein that can modulate function. Often, these additional secondary binding sites appear separate to the primary binding site, which, for example for an enzyme, may bind a substrate. In previous work, we have uncovered several examples in which secondary binding sites were discovered on proteins using fragment screening approaches. In each case, we were able to establish that the newly identified secondary binding site was biologically relevant as it was able to modulate function by the binding of a small molecule. In this study, we investigate how often secondary binding sites are located on proteins by analyzing 24 protein targets for which we have performed a fragment screen using X-ray crystallography. Our analysis shows that, surprisingly, the majority of proteins contain secondary binding sites based on their ability to bind fragments. Furthermore, sequence analysis of these previously unknown sites indicate high conservation, which suggests that they may have a biological function, perhaps via an allosteric mechanism. Comparing the physicochemical properties of the secondary sites with known primary ligand binding sites also shows broad similarities indicating that many of the secondary sites may be druggable in nature with small molecules that could provide new opportunities to modulate potential therapeutic targets.

Potential and limitations of ensemble docking.

A major problem in structure-based virtual screening applications is the appropriate selection of a single or even multiple protein structures to be used in the virtual screening process. A priori it is unknown which protein structure(s) will perform best in a virtual screening experiment. We investigated the performance of ensemble docking, as a function of ensemble size, for eight targets of pharmaceutical interest. Starting from single protein structure docking results, for each ensemble size up to 500,000 combinations of protein structures were generated, and, for each ensemble, pose prediction and virtual screening results were derived. Comparison of single to multiple protein structure results suggests improvements when looking at the performance of the worst and the average over all single protein structures to the performance of the worst and average over all protein ensembles of size two or greater, respectively. We identified several key factors affecting ensemble docking performance, including the sampling accuracy of the docking algorithm, the choice of the scoring function, and the similarity of database ligands to the cocrystallized ligands of ligand-bound protein structures in an ensemble. Due to these factors, the prospective selection of optimum ensembles is a challenging task, shown by a reassessment of published ensemble selection protocols.

Generation of Quantum Configurational Ensembles Using Approximate Potentials.

Conformational analysis is of paramount importance in drug design: it is crucial to determine pharmacological properties, understand molecular recognition processes, and characterize the conformations of ligands when unbound. Molecular Mechanics (MM) simulation methods, such as Monte Carlo (MC) and molecular dynamics (MD), are usually employed to generate ensembles of structures due to their ability to extensively sample the conformational space of molecules. The accuracy of these MM-based schemes strongly depends on the functional form of the force field (FF) and its parametrization, components that often hinder their performance. High-level methods, such as ab initio MD, provide reliable structural information but are still too computationally expensive to allow for extensive sampling. Therefore, to overcome these limitations, we present a multilevel MC method that is capable of generating quantum configurational ensembles while keeping the computational cost at a minimum. We show that FF reparametrization is an efficient route to generate FFs that reproduce QM results more closely, which, in turn, can be used as low-cost models to achieve the gold standard QM accuracy. We demonstrate that the MC acceptance rate is strongly correlated with various phase space overlap measurements and that it constitutes a robust metric to evaluate the similarity between the MM and QM levels of theory. As a more advanced application, we present a self-parametrizing version of the algorithm, which combines sampling and FF parametrization in one scheme, and apply the methodology to generate the QM/MM distribution of a ligand in aqueous solution.

Fragment-Guided Discovery of Pyrazole Carboxylic Acid Inhibitors of the Kelch-like ECH-Associated Protein 1: Nuclear Factor Erythroid 2 Related Factor 2 (KEAP1:NRF2) Protein-Protein Interaction.

The NRF2-mediated cytoprotective response is central to cellular homoeostasis, and there is increasing interest in developing small-molecule activators of this pathway as therapeutics for diseases involving chronic oxidative stress. The protein KEAP1, which regulates NRF2, is a key point for pharmacological intervention, and we recently described the use of fragment-based drug discovery to develop a tool compound that directly disrupts the protein-protein interaction between NRF2 and KEAP1. We now present the identification of a second, chemically distinct series of KEAP1 inhibitors, which provided an alternative chemotype for lead optimization. Pharmacophoric information from our original fragment screen was used to identify new hit matter through database searching and to evolve this into a new lead with high target affinity and cell-based activity. We highlight how knowledge obtained from fragment-based approaches can be used to focus additional screening campaigns in order to de-risk projects through the rapid identification of novel chemical series.

Practical High-Quality Electrostatic Potential Surfaces for Drug Discovery Using a Graph-Convolutional Deep Neural Network.

Inspecting protein and ligand electrostatic potential (ESP) surfaces in order to optimize electrostatic complementarity is a key activity in drug design. These ESP surfaces need to reflect the true electrostatic nature of the molecules, which typically means time-consuming high-level quantum mechanics (QM) calculations are required. For interactive design much faster alternative methods are required. Here, we present a graph convolutional deep neural network (DNN) model, trained on ESP surfaces derived from high quality QM calculations, that generates ESP surfaces for ligands in a fraction of a second. Additionally, we describe a method for constructing fast QM-trained ESP surfaces for proteins. We show that the DNN model generates ESP surfaces that are in good agreement with QM and that the ESP values correlate well with experimental properties relevant to medicinal chemistry. We believe that these high-quality, interactive ESP surfaces form a powerful tool for driving drug discovery programs forward. The trained model and associated code are available from https://github.com/AstexUK/ESP_DNN.

Fragment-based screening using X-ray crystallography and NMR spectroscopy.

Approaches which start from a study of the interaction of very simple molecules (fragments) with the protein target are proving to be valuable additions to drug design. Fragment-based screening allows the complementarity between a protein active site and drug-like molecules to be rapidly and effectively explored, using structural methods. Recent improvements in the intensities of laboratory X-ray sources permits the collection of greater amounts of high-quality diffraction data and have been matched by developments in automation, crystallisation and data analysis. Developments in NMR screening, including the use of cryogenically cooled NMR probes and (19)F-containing reporter molecules have expanded the scope of this technique, while increasing the availability of binding site and quantitative affinity data for the fragments. Application of these methods has led to a greater knowledge of the chemical variety, structural features and energetics of protein-fragment interactions. While fragment-based screening has already been shown to reduce the timescales of the drug discovery process, a more detailed characterisation of fragment screening hits can reveal unexpected similarities between fragment chemotypes and protein active sites leading to improved understanding of the pharmacophores and the re-use of this information against other protein targets.

A structural comparison of inhibitor binding to PKB, PKA and PKA-PKB chimera.

Although the crystal structure of the anti-cancer target protein kinase B (PKBbeta/Akt-2) has been useful in guiding inhibitor design, the closely related kinase PKA has generally been used as a structural mimic due to its facile crystallization with a range of ligands. The use of PKB-inhibitor crystallography would bring important benefits, including a more rigorous understanding of factors dictating PKA/PKB selectivity, and the opportunity to validate the utility of PKA-based surrogates. We present a "back-soaking" method for obtaining PKBbeta-ligand crystal structures, and provide a structural comparison of inhibitor binding to PKB, PKA, and PKA-PKB chimera. One inhibitor presented here exhibits no PKB/PKA selectivity, and the compound adopts a similar binding mode in all three systems. By contrast, the PKB-selective inhibitor A-443654 adopts a conformation in PKB and PKA-PKB that differs from that with PKA. We provide a structural explanation for this difference, and highlight the ability of PKA-PKB to mimic the true PKB binding mode in this case.

Identification of 4-(4-aminopiperidin-1-yl)-7H-pyrrolo[2,3-d]pyrimidines as selective inhibitors of protein kinase B through fragment elaboration.

Fragment-based screening identified 7-azaindole as a protein kinase B inhibitor scaffold. Fragment elaboration using iterative crystallography of inhibitor-PKA-PKB chimera complexes efficiently guided improvements in the potency and selectivity of the compounds, resulting in the identification of nanomolar 6-(piperidin-1-yl)purine, 4-(piperidin-1-yl)-7-azaindole, and 4-(piperidin-1-yl)pyrrolo[2,3- d]pyrimidine inhibitors of PKBbeta with antiproliferative activity and showing pathway inhibition in cells. A divergence in the binding mode was seen between 4-aminomethylpiperidine and 4-aminopiperidine containing molecules. Selectivity for PKB vs PKA was observed with 4-aminopiperidine derivatives, and the most PKB-selective inhibitor (30-fold) showed significantly different bound conformations between PKA and PKA-PKB chimera.

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

A procedure for analyzing and classifying publicly available crystal structures has been developed. It has been used to identify high-resolution protein-ligand complexes that can be assessed by reconstructing the electron density for the ligand using the deposited structure factors. The complexes have been clustered according to the protein sequences, and clusters have been discarded if they do not represent proteins thought to be of direct interest to the pharmaceutical or agrochemical industry. Rules have been used to exclude complexes containing non-drug-like ligands. One complex from each cluster has been selected where a structure of sufficient quality was available. The final Astex diverse set contains 85 diverse, relevant protein-ligand complexes, which have been prepared in a format suitable for docking and are to be made freely available to the entire research community (http://www.ccdc.cam.ac.uk). The performance of the docking program GOLD against the new set is assessed using a variety of protocols. Relatively unbiased protocols give success rates of approximately 80% for redocking into native structures, but it is possible to get success rates of over 90% with some protocols.

Identification of inhibitors of protein kinase B using fragment-based lead discovery.

Using fragment-based screening techniques, 5-methyl-4-phenyl-1H-pyrazole (IC50 80 microM) was identified as a novel, low molecular weight inhibitor of protein kinase B (PKB). Herein we describe the rapid elaboration of highly potent and ligand efficient analogues using a fragment growing approach. Iterative structure-based design was supported by protein-ligand structure determinations using a PKA-PKB "chimera" and a final protein-ligand structure of a lead compound in PKBbeta itself.

The Fragment Network: A Chemistry Recommendation Engine Built Using a Graph Database.

The hit validation stage of a fragment-based drug discovery campaign involves probing the SAR around one or more fragment hits. This often requires a search for similar compounds in a corporate collection or from commercial suppliers. The Fragment Network is a graph database that allows a user to efficiently search chemical space around a compound of interest. The result set is chemically intuitive, naturally grouped by substitution pattern and meaningfully sorted according to the number of observations of each transformation in medicinal chemistry databases. This paper describes the algorithms used to construct and search the Fragment Network and provides examples of how it may be used in a drug discovery context.

Predicting "Hot" and "Warm" Spots for Fragment Binding.

Computational fragment mapping methods aim to predict hotspots on protein surfaces where small fragments will bind. Such methods are popular for druggability assessment as well as structure-based design. However, to date researchers developing or using such tools have had no clear way of assessing the performance of these methods. Here, we introduce the first diverse, high quality validation set for computational fragment mapping. The set contains 52 diverse examples of fragment binding "hot" and "warm" spots from the Protein Data Bank (PDB). Additionally, we describe PLImap, a novel protocol for fragment mapping based on the Protein-Ligand Informatics force field (PLIff). We evaluate PLImap against the new fragment mapping test set, and compare its performance to that of simple shape-based algorithms and fragment docking using GOLD. PLImap is made publicly available from https://bitbucket.org/AstexUK/pli .

Protein-ligand binding affinity predictions by implicit solvent simulations: a tool for lead optimization?

Continuum electrostatics is combined with rigorous free-energy calculations in an effort to deliver a reliable and efficient method for in silico lead optimization. The methodology is tested by calculation of the relative binding free energies of a set of inhibitors of neuraminidase, cyclooxygenase2, and cyclin-dependent kinase 2. The calculated free energies are compared to the results obtained with explicit solvent simulations and empirical scoring functions. For cyclooxygenase2, deficiencies in the continuum electrostatics theory are identified and corrected with a modified simulation protocol. For neuraminidase, it is shown that a continuum representation of the solvent leads to markedly different protein-ligand interactions compared to the explicit solvent simulations, and a reconciliation of the two protocols is problematic. Cyclin-dependent kinase 2 proves more challenging, and none of the methods employed in this study yield high quality predictions. Despite the differences observed, for these systems, the use of an implicit solvent framework to predict the ranking of congeneric inhibitors to a protein is shown to be faster, as accurate or more accurate than the explicit solvent protocol, and superior to empirical scoring schemes.

Protein-Ligand Informatics Force Field (PLIff): Toward a Fully Knowledge Driven "Force Field" for Biomolecular Interactions.

The Protein Data Bank (PDB) contains a wealth of data on nonbonded biomolecular interactions. If this information could be distilled down to nonbonded interaction potentials, these would have some key advantages over standard force fields. However, there are some important outstanding issues to address in order to do this successfully. This paper introduces the protein-ligand informatics "force field", PLIff, which begins to address these key challenges ( https://bitbucket.org/AstexUK/pli ). As a result of their knowledge-based nature, the next-generation nonbonded potentials that make up PLIff automatically capture a wide range of interaction types, including special interactions that are often poorly described by standard force fields. We illustrate how PLIff may be used in structure-based design applications, including interaction fields, fragment mapping, and protein-ligand docking. PLIff performs at least as well as state-of-the art scoring functions in terms of pose predictions and ranking compounds in a virtual screening context.

General and targeted statistical potentials for protein-ligand interactions.

We present a novel atom-atom potential derived from a database of protein-ligand complexes. First, we clarify the similarities and differences between two statistical potentials described in the literature, PMF and Drugscore. We highlight shortcomings caused by an important factor unaccounted for in their reference states, and describe a new potential, which we name the Astex Statistical Potential (ASP). ASP's reference state considers the difference in exposure of protein atom types towards ligand binding sites. We show that this new potential predicts binding affinities with an accuracy similar to that of Goldscore and Chemscore. We investigate the influence of the choice of reference state by constructing two additional statistical potentials that differ from ASP only in this respect. The reference states in these two potentials are defined along the lines of Drugscore and PMF. In docking experiments, the potential using the new reference state proposed for ASP gives better success rates than when these literature reference states were used; a success rate similar to the established scoring functions Goldscore and Chemscore is achieved with ASP. This is the case both for a large, general validation set of protein-ligand structures and for small test sets of actives against four pharmaceutically relevant targets. Virtual screening experiments for these targets show less discrimination between the different reference states in terms of enrichment. In addition, we describe how statistical potentials can be used in the construction of targeted scoring functions. Examples are given for cdk2, using four different targeted scoring functions, biased towards increasingly large target-specific databases. Using these targeted scoring functions, docking success rates as well as enrichments are significantly better than for the general ASP scoring function. Results improve with the number of structures used in the construction of the target scoring functions, thus illustrating that these targeted ASP potentials can be continuously improved as new structural data become available.

Prediction of binding modes for ligands in the cytochromes P450 and other heme-containing proteins.

The cytochromes P450 (P450s) are a family of heme-containing monooxygenase enzymes involved in a variety of functions, including the metabolism of endogenous and exogenous substances in the human body. During lead optimization, and in drug development, many potential drug candidates are rejected because of the affinity they display for drug-metabolising P450s. Recently, crystal structures of human enzymes involved in drug metabolism have been determined, significantly augmenting the prospect of using structure-based design to modulate the binding and metabolizing properties of compounds against P450 proteins. An important step in the application of structure-based metabolic optimization is the accurate prediction of docking modes in heme binding proteins. In this paper we assess the performance of the docking program GOLD at predicting the binding mode of 45 heme-containing complexes. We achieved success rates of 64% and 57% for Chemscore and Goldscore respectively; these success rates are significantly lower than the value of 79% observed with both scoring functions for the full GOLD validation set. Re-parameterization of metal-acceptor interactions and lipophilicity of planar nitrogen atoms in the scoring functions resulted in a significant increase in the percentage of successful dockings against the heme binding proteins (Chemscore 73%, Goldscore 65%). The modified scoring functions will be useful in docking applications on P450 enzymes and other heme binding proteins.

Modeling water molecules in protein-ligand docking using GOLD.

We implemented a novel approach to score water mediation and displacement in the protein-ligand docking program GOLD. The method allows water molecules to switch on and off and to rotate around their three principal axes. A constant penalty, sigma(p), representing the loss of rigid-body entropy, is added for water molecules that are switched on, hence rewarding water displacement. We tested the methodology in an extensive validation study. First, sigma(p) is optimized against a training set of 58 protein-ligand complexes. For this training set, our algorithm correctly predicts water mediation/displacement in approximately 92% of the cases. We observed small improvements in the quality of the predicted binding modes for water-mediated complexes. In the second part of this work, an entirely independent set of 225 complexes is used. For this test set, our algorithm correctly predicts water mediation/displacement in approximately 93% of the cases. Improvements in binding mode quality were observed for individual water-mediated complexes.

Improved protein-ligand docking using GOLD.

The Chemscore function was implemented as a scoring function for the protein-ligand docking program GOLD, and its performance compared to the original Goldscore function and two consensus docking protocols, "Goldscore-CS" and "Chemscore-GS," in terms of docking accuracy, prediction of binding affinities, and speed. In the "Goldscore-CS" protocol, dockings produced with the Goldscore function are scored and ranked with the Chemscore function; in the "Chemscore-GS" protocol, dockings produced with the Chemscore function are scored and ranked with the Goldscore function. Comparisons were made for a "clean" set of 224 protein-ligand complexes, and for two subsets of this set, one for which the ligands are "drug-like," the other for which they are "fragment-like." For "drug-like" and "fragment-like" ligands, the docking accuracies obtained with Chemscore and Goldscore functions are similar. For larger ligands, Goldscore gives superior results. Docking with the Chemscore function is up to three times faster than docking with the Goldscore function. Both combined docking protocols give significant improvements in docking accuracy over the use of the Goldscore or Chemscore function alone. "Goldscore-CS" gives success rates of up to 81% (top-ranked GOLD solution within 2.0 A of the experimental binding mode) for the "clean list," but at the cost of long search times. For most virtual screening applications, "Chemscore-GS" seems optimal; search settings that give docking speeds of around 0.25-1.3 min/compound have success rates of about 78% for "drug-like" compounds and 85% for "fragment-like" compounds. In terms of producing binding energy estimates, the Goldscore function appears to perform better than the Chemscore function and the two consensus protocols, particularly for faster search settings. Even at docking speeds of around 1-2 min/compound, the Goldscore function predicts binding energies with a standard deviation of approximately 10.5 kJ/mol.

A new test set for validating predictions of protein-ligand interaction.

We present a large test set of protein-ligand complexes for the purpose of validating algorithms that rely on the prediction of protein-ligand interactions. The set consists of 305 complexes with protonation states assigned by manual inspection. The following checks have been carried out to identify unsuitable entries in this set: (1) assessing the involvement of crystallographically related protein units in ligand binding; (2) identification of bad clashes between protein side chains and ligand; and (3) assessment of structural errors, and/or inconsistency of ligand placement with crystal structure electron density. In addition, the set has been pruned to assure diversity in terms of protein-ligand structures, and subsets are supplied for different protein-structure resolution ranges. A classification of the set by protein type is available. As an illustration, validation results are shown for GOLD and SuperStar. GOLD is a program that performs flexible protein-ligand docking, and SuperStar is used for the prediction of favorable interaction sites in proteins. The new CCDC/Astex test set is freely available to the scientific community (http://www.ccdc.cam.ac.uk).