Computer-aided drug design at Boehringer Ingelheim.

Computer-Aided Drug Design (CADD) is an integral part of the drug discovery endeavor at Boehringer Ingelheim (BI). CADD contributes to the evaluation of new therapeutic concepts, identifies small molecule starting points for drug discovery, and develops strategies for optimizing hit and lead compounds. The CADD scientists at BI benefit from the global use and development of both software platforms and computational services. A number of computational techniques developed in-house have significantly changed the way early drug discovery is carried out at BI. In particular, virtual screening in vast chemical spaces, which can be accessed by combinatorial chemistry, has added a new option for the identification of hits in many projects. Recently, a new framework has been implemented allowing fast, interactive predictions of relevant on and off target endpoints and other optimization parameters. In addition to the introduction of this new framework at BI, CADD has been focusing on the enablement of medicinal chemists to independently perform an increasing amount of molecular modeling and design work. This is made possible through the deployment of MOE as a global modeling platform, allowing computational and medicinal chemists to freely share ideas and modeling results. Furthermore, a central communication layer called the computational chemistry framework provides broad access to predictive models and other computational services.

Rodent selectivity of piperidine-4-yl-1H-indoles, a series of CC chemokine receptor-3 (CCR3) antagonists: insights from a receptor model.

Rodent selectivity data of piperidine-4-yl-1H-indoles, a series of CC chemokine receptor-3 (CCR3) antagonists, are presented and discussed as part of an overall optimization effort within this lead compound class. Although attachment of an acidic moiety to the 1-position of the indole led to an overall balanced in vitro profile, in particular reducing inhibition of the hERG channel, potency on the rat and mouse receptor worsened. These findings could be rationalized in the context of a CCR3 homology model.

The impact of data integrity on decision making in early lead discovery.

Data driven decision making is a key element of today's pharmaceutical research, including early drug discovery. It comprises questions like which target to pursue, which chemical series to pursue, which compound to make next, or which compound to select for advanced profiling and promotion to pre-clinical development. In the following paper we will exemplify how data integrity, i.e. the context data is generated in and auxiliary information that is provided for individual result records, can influence decision making in early lead discovery programs. In addition we will describe some approaches which we pursue at Boehringer Ingelheim to reduce the risk for getting misguided.

Virtual clinical QT exposure-response studies - A translational computational approach.

BACKGROUND AND PURPOSE: A recent paradigm shift in proarrhythmic risk assessment suggests that the integration of clinical, non-clinical, and computational evidence can be used to reach a comprehensive understanding of the proarrhythmic potential of drug candidates. While current computational methodologies focus on predicting the incidence of proarrhythmic events after drug administration, the objective of this study is to predict concentration-response relationships of QTc as a clinical endpoint. EXPERIMENTAL APPROACH: Full heart computational models reproducing human cardiac populations were created to predict the concentration-response relationship of changes in the QT interval as recommended for clinical trials. The concentration-response relationship of the QT-interval prolongation obtained from the computational cardiac population was compared against the relationship from clinical trial data for a set of well-characterized compounds: moxifloxacin, dofetilide, verapamil, and ondansetron. KEY RESULTS: Computationally derived concentration-response relationships of QT interval changes for three of the four drugs had slopes within the confidence interval of clinical trials (dofetilide, moxifloxacin and verapamil) when compared to placebo-corrected concentration-ΔQT and concentration-ΔQT regressions. Moxifloxacin showed a higher intercept, outside the confidence interval of the clinical data, demonstrating that in this example, the standard linear regression does not appropriately capture the concentration-response results at very low concentrations. The concentrations corresponding to a mean QTc prolongation of 10 ms were consistently lower in the computational model than in clinical data. The critical concentration varied within an approximate ratio of 0.5 (moxifloxacin and ondansetron) and 1 times (dofetilide, verapamil) the critical concentration observed in human clinical trials. Notably, no other in silico methodology can approximate the human critical concentration values for a QT interval prolongation of 10 ms. CONCLUSION AND IMPLICATIONS: Computational concentration-response modelling of a virtual population of high-resolution, 3-dimensional cardiac models can provide comparable information to clinical data and could be used to complement pre-clinical and clinical safety packages. It provides access to an unlimited exposure range to support trial design and can improve the understanding of pre-clinical-clinical translation.

Mining electronic laboratory notebooks: analysis, retrosynthesis, and reaction based enumeration.

An approach to automatically analyze and use the knowledge contained in electronic laboratory notebooks (ELNs) has been developed. Reactions were reduced to their reactive center and converted to a string representation (SMIRKS) which formed the basis for reaction classification and in silico (retro-)synthesis. Of the SMIRKS that occurred at least five times, 98% successfully regenerated the original product. The extracted reaction rules (SMIRKS) and corresponding reactants span a virtual chemical space which showed a strong dependence on the size of the reactive center. Whereas relatively few robust reaction types were sufficient to describe a large part of all reactions, considerably more reaction rules were necessary to cover all reactions in the ELN. Furthermore, reaction sequences were extracted to identify frequent combinations and diversifying reaction steps. Based on the extracted knowledge a (retro-)synthesis tool was built allowing for de novo design of compounds which have a high chance of being synthetically accessible. In an example application of the de novo design tool, various feasible retrosynthetic routes to the query molecule were obtained. Reaction based enumeration along the top ranked route yielded a library of 29 920 compounds with diverse properties, 99.9% of which are novel in the sense that they are unknown to the public domain.

Self-organizing fuzzy graphs for structure-based comparison of protein pockets.

Patterns of receptor-ligand interaction can be conserved in functionally equivalent proteins even in the absence of sequence homology. Therefore, structural comparison of ligand-binding pockets and their pharmacophoric features allow for the characterization of so-called "orphan" proteins with known three-dimensional structure but unknown function, and predict ligand promiscuity of binding pockets. We present an algorithm for rapid pocket comparison (PoLiMorph), in which protein pockets are represented by self-organizing graphs that fill the volume of the cavity. Vertices in these three-dimensional frameworks contain information about the local ligand-receptor interaction potential coded by fuzzy property labels. For framework matching, we developed a fast heuristic based on the maximum dispersion problem, as an alternative to techniques utilizing clique detection or geometric hashing algorithms. A sophisticated scoring function was applied that incorporates knowledge about property distributions and ligand-receptor interaction patterns. In an all-against-all virtual screening experiment with 207 pocket frameworks extracted from a subset of PDBbind, PoLiMorph correctly assigned 81% of 69 distinct structural classes and demonstrated sustained ability to group pockets accommodating the same ligand chemotype. We determined a score threshold that indicates "true" pocket similarity with high reliability, which not only supports structure-based drug design but also allows for sequence-independent studies of the proteome.

Architectural repertoire of ligand-binding pockets on protein surfaces.

Knowledge of the three-dimensional structure of ligand binding sites in proteins provides valuable information for computer-assisted drug design. We present a method for the automated extraction and classification of ligand binding site topologies, in which protein surface cavities are represented as branched frameworks. The procedure employs a growing neural gas approach for pocket topology assignment and pocket framework generation. We assessed the structural diversity of 623 known ligand binding site topologies based on framework cluster analysis. At a resolution of 5 A only 23 structurally distinct topology groups were formed; this suggests an overall limited structural diversity of ligand-accommodating protein cavities. Higher resolution allowed for identification of protein-family specific pocket features. Pocket frameworks highlight potentially preferred modes of ligand-receptor interactions and will help facilitate the identification of druggable subpockets suitable for ligand affinity and selectivity optimization.

Form follows function: shape analysis of protein cavities for receptor-based drug design.

Identification of potential ligand-binding pockets is an initial step in receptor-based drug design. While many geometric or energy-based binding-site prediction methods characterize the size and shape of protein cavities, few of them offer an estimate of the pocket's ability to bind small drug-like molecules. Here, we present a shape-based technique to examine binding-site druggability from the crystal structure of a given protein target. The method includes the PocketPicker algorithm to determine putative binding-site volumes for ligand-interaction. Pocket shape descriptors were calculated for both known ligand binding sites and empty pockets and were then subjected to self-organizing map clustering. Descriptors were calculated for structures derived from a database of representative drug-protein complexes with experimentally determined binding affinities to characterize the "druggable pocketome". The new method provides a means for selecting drug targets and potential ligand-binding pockets based on structural considerations and addresses orphan binding sites.

Applications of Deep-Learning in Exploiting Large-Scale and Heterogeneous Compound Data in Industrial Pharmaceutical Research.

In recent years, the development of high-throughput screening (HTS) technologies and their establishment in an industrialized environment have given scientists the possibility to test millions of molecules and profile them against a multitude of biological targets in a short period of time, generating data in a much faster pace and with a higher quality than before. Besides the structure activity data from traditional bioassays, more complex assays such as transcriptomics profiling or imaging have also been established as routine profiling experiments thanks to the advancement of Next Generation Sequencing or automated microscopy technologies. In industrial pharmaceutical research, these technologies are typically established in conjunction with automated platforms in order to enable efficient handling of screening collections of thousands to millions of compounds. To exploit the ever-growing amount of data that are generated by these approaches, computational techniques are constantly evolving. In this regard, artificial intelligence technologies such as deep learning and machine learning methods play a key role in cheminformatics and bio-image analytics fields to address activity prediction, scaffold hopping, de novo molecule design, reaction/retrosynthesis predictions, or high content screening analysis. Herein we summarize the current state of analyzing large-scale compound data in industrial pharmaceutical research and describe the impact it has had on the drug discovery process over the last two decades, with a specific focus on deep-learning technologies.

Machine learning techniques for in silico modeling of drug metabolism.

The computational assessment of drug metabolism has gained considerable interest in pharmaceutical research. Amongst others, machine learning techniques have been employed to model relationships between the chemical structure of a compound and its metabolic fate. Examples for these techniques, which were originally developed in fields far from drug discovery, are artificial neural networks or support vector machines. This paper summarizes the application of various machine learning techniques to predict the interaction between organic molecules and metabolic enzymes. More complex endpoints such as metabolic stability or in vivo clearance will also be addressed. It is shown that the prediction of metabolic endpoints with machine learning techniques has made considerable progress over the past few years. Depending on the procedure used, either classification or quantitative prediction is possible for even large and diverse compound sets. Together with the expanding experimental data basis, these in silico methods have become valuable tools in the drug discovery and development process.

Designing Multi-target Compound Libraries with Gaussian Process Models.

We present the application of machine learning models to selecting G protein-coupled receptor (GPCR)-focused compound libraries. The library design process was realized by ant colony optimization. A proprietary Boehringer-Ingelheim reference set consisting of 3519 compounds tested in dose-response assays at 11 GPCR targets served as training data for machine learning and activity prediction. We compared the usability of the proprietary data with a public data set from ChEMBL. Gaussian process models were trained to prioritize compounds from a virtual combinatorial library. We obtained meaningful models for three of the targets (5-HT2c , MCH, A1), which were experimentally confirmed for 12 of 15 selected and synthesized or purchased compounds. Overall, the models trained on the public data predicted the observed assay results more accurately. The results of this study motivate the use of Gaussian process regression on public data for virtual screening and target-focused compound library design.

Multivariate modeling of cytochrome P450 3A4 inhibition.

In the early phases of current pharmaceutical research projects, huge numbers of compounds are tested on their biological activity with respect to a certain target by experimental or virtual screening campaigns. To reduce the attrition rate in later stages of a project, other relevant properties such as physicochemical and ADMET (absorption, distribution, metabolism, excretion, toxicity) properties should be assessed as early as possible in lead discovery and optimization. The present study describes the development of in silico models to predict the inhibition of human cytochrome P450 3A4 (CYP3A4) from calculated molecular descriptors. The models were trained and validated using a set of 967 structural diverse drug-like research compounds with an experimentally determined CYP3A4 inhibition potency (IC50 value) which was carefully split into a training and a test set. For classification models, the data sets were further subdivided into strong, medium, and weak inhibitors. Different descriptor sets were used to cover various aspects of molecular properties, including properties derived from the 2D structure, the interaction of the molecule with its environment, and properties derived from quantum-mechanical calculations. The descriptors were related to the CYP3A4 inhibition potency by multivariate data analysis methods such as partial least-squares projection to latent structures (PLS), PLS discriminant analysis (PLS-DA), and soft independent class modeling (SIMCA). The squared correlation between experimental and predicted IC50 values of the previously unseen test set compounds was Qext2=0.6 for the best PLS models, corresponding to a root mean squared error (RMSE) of RMSE=0.45 (logarithm of IC50). The best PLS-DA models were able to correctly classify more than 60% of the test set compounds, whereas almost no strong inhibitors were wrongly classified as weak inhibitors and vice versa. Furthermore, relevant molecular properties were identified which are closely related to the CYP3A4 inhibition potency of a compound. The results presented here are very encouraging since our models could, for instance, serve to flag problematic compounds or to guide further synthesis efforts.

What can we learn from molecular dynamics simulations for GPCR drug design?

Recent years have seen a tremendous progress in the elucidation of experimental structural information for G-protein coupled receptors (GPCRs). Although for the vast majority of pharmaceutically relevant GPCRs structural information is still accessible only by homology models the steadily increasing amount of structural information fosters the application of structure-based drug design tools for this important class of drug targets. In this article we focus on the application of molecular dynamics (MD) simulations in GPCR drug discovery programs. Typical application scenarios of MD simulations and their scope and limitations will be described on the basis of two selected case studies, namely the binding of small molecule antagonists to the human CC chemokine receptor 3 (CCR3) and a detailed investigation of the interplay between receptor dynamics and solvation for the binding of small molecules to the human muscarinic acetylcholine receptor 3 (hM3R).

Computational approaches to predict drug metabolism.

BACKGROUND: Metabolism is one of the key parameters to be investigated and optimized in drug discovery projects. Metabolically unstable compounds or potential inhibitors of important enzymes should be detected as early as possible. As more compounds are synthesized than can be investigated experimentally, powerful computational methods are needed. OBJECTIVE: We give an overview of state-of-the-art in-silico methods to predict experimental metabolic endpoints with a focus on the applicability in pharmaceutical industry. A macroscopic as well as a microscopic view of the metabolic fate and the interaction with metabolizing enzymes are given. METHODS: Ligand-, protein- and rule-based approaches are presented. CONCLUSION: Although considerable progress has been made, the results of the calculations still need careful inspection. The domain of applicability of the models as well as methodological limitations have to be taken into account.

Bias-correction of regression models: a case study on hERG inhibition.

In the present work we develop a predictive QSAR model for the blockade of the hERG channel. Additionally, this specific end point is used as a test scenario to develop and evaluate several techniques for fusing predictions from multiple regression models. hERG inhibition models which are presented here are based on a combined data set of roughly 550 proprietary and 110 public domain compounds. Models are built using various statistical learning techniques and different sets of molecular descriptors. Single Support Vector Regression, Gaussian Process, or Random Forest models achieve root mean-squared errors of roughly 0.6 log units as determined from leave-group-out cross-validation. An analysis of the evaluation strategy on the performance estimates shows that standard leave-group-out cross-validation yields overly optimistic results. As an alternative, a clustered cross-validation scheme is introduced to obtain a more realistic estimate of the model performance. The evaluation of several techniques to combine multiple prediction models shows that the root mean squared error as determined from clustered cross-validation can be reduced from 0.73 +/- 0.01 to 0.57 +/- 0.01 using a local bias correction strategy.

A composite model for HERG blockade.

hERG blockade is one of the major toxicological problems in lead structure optimization. Reliable ligand-based in silico models for predicting hERG blockade therefore have considerable potential for saving time and money, as patch-clamp measurements are very expensive and no crystal structures of the hERG-encoded channel are available. Herein we present a predictive QSAR model for hERG blockade that differentiates between specific and nonspecific binding. Specific binders are identified by preliminary pharmacophore scanning. In addition to descriptor-based models for the compounds selected as hitting one of two different pharmacophores, we also use a model for nonspecific binding that reproduces blocking properties of molecules that do not fit either of the two pharmacophores. PLS and SVR models based on interpretable quantum mechanically derived descriptors on a literature dataset of 113 molecules reach overall R(2) values between 0.60 and 0.70 for independent validation sets and R(2) values between 0.39 and 0.76 after partitioning according to the pharmacophore search for the test sets. Our findings suggest that hERG blockade may occur through different types of binding, so that several different models may be necessary to assess hERG toxicity.

A support vector machine approach to classify human cytochrome P450 3A4 inhibitors.

The cytochrome P450 (CYP) enzyme superfamily plays a major role in the metabolism of commercially available drugs. Inhibition of these enzymes by a drug may result in a plasma level increase of another drug, thus leading to unwanted drug-drug interactions when two or more drugs are coadministered. Therefore, fast and reliable in silico methods predicting CYP inhibition from calculated molecular properties are an important tool which can be applied to assess both already synthesized as well as virtual compounds. We have studied the performance of support vector machines (SVMs) to classify compounds according to their potency to inhibit CYP3A4. The data set for model generation consists of more than 1300 structural diverse drug-like research molecules which were divided into training and test sets. The predictive power of SVMs crucially depends on a careful selection of parameters specifying the kernel function and the penalty for misclassifications. In this study we have investigated a procedure to identify a valid set of SVM parameters which is based on a sampling of the parameter space on a regular grid. From this set of parameters, either single SVMs or SVM committees were trained to distinguish between strong and weak inhibitors or to achieve a more realistic three-class assignment, with one class representing medium inhibitors. This workflow was studied for several kernel functions and descriptor sets. All SVM models performed significantly better than PLS-DA models which were generated from the corresponding descriptor sets. As a very promising result, simple two-dimensional (2D) descriptors yield a three-class model which correctly classifies more than 70% of the test set. Our work illustrates that SVMs used in combination with simple 2D descriptors provide a very effective and reliable tool which allows a fast assessment of CYP3A4 inhibition potency in an early in silico filtering process.

Probing electric fields in protein cavities by using the vibrational stark effect of carbon monoxide.

To determine the magnitude and direction of the internal electric field in the Xe4 cavity of myoglobin mutant L29W-S108L, we have studied the vibrational Stark effect of carbon monoxide (CO) using infrared spectroscopy at cryogenic temperatures. CO was photodissociated from the heme iron and deposited selectively in Xe4. Its infrared spectrum exhibits Stark splitting into two bands associated with CO in opposite orientations. Two different photoproduct states can be distinguished, C' and C'', with markedly different properties. For C', characteristic temperature-dependent changes of the area, shift, and width were analyzed, based on a dynamic model in which the CO performs fast librations within a double-well model potential. For the barrier between the wells, a height of approximately 1.8 kJ/mol was obtained, in which the CO performs oscillations at an angular frequency of approximately 25 cm(-1). The magnitude of the electric field in the C' conformation was determined as 11.1 MV/cm; it is tilted by an angle of 29 degrees to the symmetry axis of the potential. Above 140 K, a protein relaxation leads to a significantly altered photoproduct, C'', with a smaller Stark splitting and a more confining potential (barrier >4 kJ/mol) governing the CO librations.

Structural, dynamic, and energetic aspects of long-range electron transfer in photosynthetic reaction centers.

Intramolecular electron transfer within proteins plays an essential role in biological energy transduction. Electron donor and acceptor cofactors are bound in the protein matrix at specific locations, and protein-cofactor interactions as well as protein conformational changes can markedly influence the electron transfer rates. To assess these effects, we have investigated charge recombination from the primary quinone acceptor to the special pair bacteriochlorophyll dimer in wild-type reaction centers of Rhodobacter sphaeroides and four mutants with widely modified free energy gaps. After light-induced charge separation, the recombination kinetics were measured in the light- and dark-adapted forms of the protein from 10 to 300 K. The data were analyzed by using the spin-boson model, which allowed us to self-consistently determine the electronic coupling energy, the distribution of energy gaps, the spectral density of phonons, and the reorganization energy. The analysis revealed slow changes of the energy gap after charge separation. Interesting correlations of the control parameters governing electron transfer were found and related to structural and dynamic properties of the protein.