Edward W. Lowe

Computational Methods in Drug Discovery

Computer-aided drug discovery/design methods have played a major role in the development of therapeutically important small molecules for over three decades. These methods are broadly classified as either structure-based or ligand-based methods. Structure-based methods are in principle analogous to high-throughput screening in that both target and ligand structure information is imperative. Structure-based approaches include ligand docking, pharmacophore, and ligand design methods. The article discusses theory behind the most important methods and recent successful applications. Ligand-based methods use only ligand information for predicting activity depending on its similarity/dissimilarity to previously known active ligands. We review widely used ligand-based methods such as ligand-based pharmacophores, molecular descriptors, and quantitative structure-activity relationships. In addition, important tools such as target/ligand data bases, homology modeling, ligand fingerprint methods, etc., necessary for successful implementation of various computer-aided drug discovery/design methods in a drug discovery campaign are discussed. Finally, computational methods for toxicity prediction and optimization for favorable physiologic properties are discussed with successful examples from literature.

Benchmarking Ligand-Based Virtual High-Throughput Screening with the PubChem Database

With the rapidly increasing availability of High-Throughput Screening (HTS) data in the public domain, such as the PubChem database, methods for ligand-based computer-aided drug discovery (LB-CADD) have the potential to accelerate and reduce the cost of probe development and drug discovery efforts in academia. We assemble nine data sets from realistic HTS campaigns representing major families of drug target proteins for benchmarking LB-CADD methods. Each data set is public domain through PubChem and carefully collated through confirmation screens validating active compounds. These data sets provide the foundation for benchmarking a new cheminformatics framework BCL::ChemInfo, which is freely available for non-commercial use. Quantitative structure activity relationship (QSAR) models are built using Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), Decision Trees (DTs), and Kohonen networks (KNs). Problem-specific descriptor optimization protocols are assessed including Sequential Feature Forward Selection (SFFS) and various information content measures. Measures of predictive power and confidence are evaluated through cross-validation, and a consensus prediction scheme is tested that combines orthogonal machine learning algorithms into a single predictor. Enrichments ranging from 15 to 101 for a TPR cutoff of 25% are observed.

Discoidin domain receptor 1 (DDR1) kinase as target for structure-based drug discovery

Discoidin domain receptor (DDR) 1 and 2 are transmembrane receptors that belong to the family of receptor tyrosine kinases (RTK). Upon collagen binding, DDRs transduce cellular signaling involved in various cell functions, including cell adhesion, proliferation, differentiation, migration, and matrix homeostasis. Altered DDR function resulting from either mutations or overexpression has been implicated in several types of disease, including atherosclerosis, inflammation, cancer, and tissue fibrosis. Several established inhibitors, such as imatinib, dasatinib, and nilotinib, originally developed as Abelson murine leukemia (Abl) kinase inhibitors, have been found to inhibit DDR kinase activity. As we review here, recent discoveries of novel inhibitors and their co-crystal structure with the DDR1 kinase domain have made structure-based drug discovery for DDR1 amenable.

RECONSTRUCTION OF SAXS PROFILES FROM PROTEIN STRUCTURES

Small angle X-ray scattering (SAXS) is used for low resolution structural characterization of proteins often in combination with other experimental techniques. After briefly reviewing the theory of SAXS we discuss computational methods based on 1) the Debye equation and 2) Spherical Harmonics to compute intensity profiles from a particular macromolecular structure. Further, we review how these formulas are parameterized for solvent density and hydration shell adjustment. Finally we introduce our solution to compute SAXS profiles utilizing GPU acceleration.

Evidence for Substrate Preorganization in the Peptidylglycine α-Amidating Monooxygenase Reaction Describing the Contribution of Ground State Structure to Hydrogen Tunneling

Peptidylglycine α-amidating monooxygenase (PAM) is a bifunctional enzyme which catalyzes the post-translational modification of inactive C-terminal glycine-extended peptide precursors to the corresponding bioactive α-amidated peptide hormone. This conversion involves two sequential reactions both of which are catalyzed by the separate catalytic domains of PAM. The first step, the copper-, ascorbate-, and O(2)-dependent stereospecific hydroxylation at the α-carbon of the C-terminal glycine, is catalyzed by peptidylglycine α-hydroxylating monooxygenase (PHM). The second step, the zinc-dependent dealkylation of the carbinolamide intermediate, is catalyzed by peptidylglycine amidoglycolate lyase. Quantum mechanical tunneling dominates PHM-dependent C(α)-H bond activation. This study probes the substrate structure dependence of this chemistry using a set of N-acylglycine substrates of varying hydrophobicity. Primary deuterium kinetic isotope effects (KIEs), molecular mechanical docking, alchemical free energy perturbation, and equilibrium molecular dynamics were used to study the role played by ground-state substrate structure on PHM catalysis. Our data show that all Ν-acylglycines bind sequentially to PHM in an equilibrium-ordered fashion. The primary deuterium KIE displays a linear decrease with respect to acyl chain length for straight-chain N-acylglycine substrates. Docking orientation of these substrates displayed increased dissociation energy proportional to hydrophobic pocket interaction. The decrease in KIE with hydrophobicity was attributed to a preorganization event which decreased reorganization energy by decreasing the conformational sampling associated with ground state substrate binding. This is the first example of preorganization in the family of noncoupled copper monooxygenases.

Imino-oxy Acetic Acid Dealkylation as Evidence for an Inner-Sphere Alcohol Intermediate in the Reaction Catalyzed by Peptidylglycine α-Hydroxylating Monooxygenase

Peptidylglycine alpha-hydroxylating monooxygenase (PHM, EC 1.14.17.3) catalyzes the stereospecific hydroxylation of a glycyl alpha-carbon in a reaction that requires O(2) and ascorbate. Subsequent dealkylation of the alpha-hydroxyglycine by another enzyme, peptidylamidoglycolate lyase (PAL. EC 4.3.2.5), yields a bioactive amide and glyoxylate. PHM is a noncoupled, type II dicopper monooxygenase which activates O(2) at only a single copper atom, Cu(M). In this study, the PHM mechanism was probed using a non-natural substrate, benzaldehyde imino-oxy acetic acid (BIAA). PHM catalyzes the O-oxidative dealkylation of BIAA to benzaldoxime and glyoxylate with no involvement of PAL. The minimal kinetic mechanism for BIAA was shown to be steady-state ordered using primary deuterium kinetic isotope effects. The (D)(V/K)(APPARENT, BIAA) decreased from 14.7 +/- 1.0 as [O(2)] --> 0 to 1.0 +/- 0.2 as [O(2)] --> infinity suggesting the dissociation rate constant from the PHM x BIAA complex decreases as [O(2)] increases; thereby, reducing the steady-state concentration of [PHM](free). BIAA was further used to differentiate between potential oxidative Cu/O species using a QM/MM reaction coordinate simulation to determine which species could yield product O-dealkylation that matched our experimental data. The results of this study provided compelling evidence for the presence of a covalently linked Cu(II)-alkoxide intermediate with a quartet spin state responsible BIAA oxidation.

BCL::SAXS: GPU accelerated Debye method for computation of small angle X‐ray scattering profiles

Small angle X‐ray scattering (SAXS) is an experimental technique used for structural characterization of macromolecules in solution. Here, we introduce BCL::SAXS—an algorithm designed to replicate SAXS profiles from rigid protein models at different levels of detail. We first show our derivation of BCL::SAXS and compare our results with the experimental scattering profile of hen egg white lysozyme. Using this protein we show how to generate SAXS profiles representing: (1) complete models, (2) models with approximated side chain coordinates, and (3) models with approximated side chain and loop region coordinates. We evaluated the ability of SAXS profiles to identify a correct protein topology from a non‐redundant benchmark set of proteins. We find that complete SAXS profiles can be used to identify the correct protein by receiver operating characteristic (ROC) analysis with an area under the curve (AUC) > 99%. We show how our approximation of loop coordinates between secondary structure elements improves protein recognition by SAχS for protein models without loop regions and side chains. Agreement with SAXS data is a necessary but not sufficient condition for structure determination. We conclude that experimental SAXS data can be used as a filter to exclude protein models with large structural differences from the native. Proteins 2015; 83:1500–1512.

GPU-accelerated machine learning techniques enable QSAR modeling of large HTS data

Quantitative structure activity relationship (QSAR) modeling using high-throughput screening (HTS) data is a powerful technique which enables the construction of predictive models. These models are utilized for the in silico screening of libraries of molecules for which experimental screening methods are both cost- and time-expensive. Machine learning techniques excel in QSAR modeling where the relationship between structure and activity is often complex and non-linear. As these HTS data sets continue to increase in number of compounds screened, extensive feature selection and cross validation becomes computationally expensive. Leveraging massively parallel architectures such as graphics processing units (GPUs) to accelerate the training algorithms for these machine learning techniques is a cost-efficient manner in which to combat this problem. In this work, several machine learning techniques are ported in OpenCL for GPU-acceleration to enable construction of QSAR ensemble models using HTS data. We report computational performance numbers using several HTS data sets freely available from PubChem database. We also report results of a case study using HTS data for a target of pharmacological and pharmaceutical relevance, cytochrome P450 3A4, for which an enrichment of 94% of the theoretical maximum is achieved.

3D-QSAR CoMFA study of benzoxazepine derivatives as mGluR5 positive allosteric modulators.

Positive allosteric modulation of the metabotropic glutamate receptor subtype 5 was studied by conducting a comparative molecular field analysis on 118 benzoxazepine derivatives. The model with the best predictive ability retained significant cross-validated correlation coefficients of q(2) = 0.58 (r(2) = 0.81) yielding a standard error of 0.20 in pEC(50) for this class of compounds. The subsequent contour maps highlight the structural features pertinent to the bioactivity values of benzoxazepines.

Mechanistic studies of the tyrosinase-catalyzed oxidative cyclocondensation of 2-aminophenol to 2-aminophenoxazin-3-one.

Tyrosinase (EC 1.14.18.1) catalyzes the monophenolase and diphenolase reaction associated with vertebrate pigmentation and fruit/vegetable browning. Tyrosinase is an oxygen-dependent, dicopper enzyme that has three states: Emet, Eoxy, and Edeoxy. The diphenolase activity can be carried out by both the met and the oxy states of the enzyme while neither mono- nor diphenolase activity results from the deoxy state. In this study, the oxidative cyclocondensation of 2-aminophenol (OAP) to the corresponding 2-aminophenoxazin-3-one (APX) by mushroom tyrosinase was investigated. Using a combination of various steady- and pre-steady state methodologies, we have investigated the kinetic and chemical mechanism of this reaction. The kcat for OAP is 75 ± 2s(-1), K(OAP)M = 1.8 ± 0.2mM, K(O2)M =25 ± 4 μM with substrates binding in a steady-state preferred fashion. Stopped flow and global analysis support a model where OAP preferentially binds to the oxy form over the met (k7 ≫ k1). For the met form, His269 and His61 are the proposed bases, while the oxy form uses the copper-peroxide and His61 for the sequential deprotonation of anilinic and phenolic hydrogens. Solvent KIEs show proton transfer to be increasingly rate limiting for kcat/K(OAP)M as [O2] → 0 μM (1.38 ± 0.06) decreasing to 0.83 ± 0.03 as [O2] → ∞ reflecting a partially rate limiting μ-OH bond cleavage (E met) and formation (E oxy) following protonation in the transition state. The coupling and cyclization reactions of o-quinone imine and OAP pass through a phenyliminocyclohexadione intermediate to APX, forming at a rate of 6.91 ± 0.03 μM(-1)s(-1) and 2.59E-2 ± 5.31E-4s(-1). Differences in reactivity attributed to the anilinic moiety of OAP with o-diphenols are discussed.