Active Learning Approach for Guiding Site-of-Metabolism Measurement and Annotation.

The ability to determine and predict metabolically labile atom positions in a molecule (also called "sites of metabolism" or "SoMs") is of high interest to the design and optimization of bioactive compounds, such as drugs, agrochemicals, and cosmetics. In recent years, several in silico models for SoM prediction have become available, many of which include a machine-learning component. The bottleneck in advancing these approaches is the coverage of distinct atom environments and rare and complex biotransformation events with high-quality experimental data. Pharmaceutical companies typically have measured metabolism data available for several hundred to several thousand compounds. However, even for metabolism experts, interpreting these data and assigning SoMs are challenging and time-consuming. Therefore, a significant proportion of the potential of the existing metabolism data, particularly in machine learning, remains dormant. Here, we report on the development and validation of an active learning approach that identifies the most informative atoms across molecular data sets for SoM annotation. The active learning approach, built on a highly efficient reimplementation of SoM predictor FAME 3, enables experts to prioritize their SoM experimental measurements and annotation efforts on the most rewarding atom environments. We show that this active learning approach yields competitive SoM predictors while requiring the annotation of only 20% of the atom positions required by FAME 3. The source code of the approach presented in this work is publicly available.

Data-Driven Global Assessment of Protein Kinase Inhibitors with Emphasis on Covalent Compounds.

Large-scale analysis of public human and mouse protein kinase inhibitor (PKI) data identified more than 155,000 human PKIs (and ∼3000 murine PKIs), for which reliable activity measurements were available. Human PKIs were active against 440 kinases (85% coverage of the kinome). Over the past years, there has been substantial growth of human PKIs, dominated by inhibitors with single-kinase annotations and high core structure diversity. Human PKIs included an unexpectedly large number of nearly 14,000 covalent PKIs (CPKIs), ∼87% of which contained acrylamide or heterocyclic urea warheads. These CPKIs were active against a large number of 369 human kinases. The promiscuity of PKIs and CPKIs was overall comparable. However, there was a notable enrichment of acrylamide- but not heterocyclic urea-containing CPKIs among most promiscuous inhibitors. Furthermore, CPKIs with both warheads had significantly higher potency than structurally analogous PKIs. Taken together, these findings have several implications for medicinal chemistry that are discussed.

Evaluating the performance of machine-learning regression models for pharmacokinetic drug-drug interactions.

Combination therapy or concomitant drug administration can be associated with pharmacokinetic drug-drug interactions, increasing the risk of adverse drug events and reduced drug efficacy. Thus far, machine-learning models have been developed that can classify drug-drug interactions. However, to enable quantification of the pharmacokinetic effects of a drug-drug interaction, regression-based machine learning should be explored. Therefore, this study investigated the use of regression-based machine learning to predict changes in drug exposure caused by pharmacokinetic drug-drug interactions. Fold changes in exposure relative to substrate drug monotherapy were collected from 120 clinical drug-drug interaction studies extracted from the Washington Drug Interaction Database and SimCYP compound library files. Drug characteristics (features) were collected such as structure, physicochemical properties, in vitro pharmacokinetic properties, cytochrome P450 metabolic activity, and population characteristics. Three different regression-based supervised machine-learning models were then applied to the prediction task: random forest, elastic net, and support vector regressor. Model performance was evaluated using fivefold cross-validation. Strongest performance was observed with support vector regression, with 78% of predictions within twofold of the observed exposure changes. The results show that changes in drug exposure can be predicted with reasonable accuracy using regression-based machine-learning models trained on data available early in drug discovery. This has potential applications in enabling earlier drug-drug interaction risk assessment for new drug candidates.

Comparing the applications of machine learning, PBPK, and population pharmacokinetic models in pharmacokinetic drug-drug interaction prediction.

The gold-standard approach for modeling pharmacokinetic mediated drug-drug interactions is the use of physiologically-based pharmacokinetic modeling and population pharmacokinetics. However, these models require extensive amounts of drug-specific data generated from a wide variety of in vitro and in vivo models, which are later refined with clinical data and system-specific parameters. Machine learning has the potential to be utilized for the prediction of drug-drug interactions much earlier in the drug discovery cycle, using inputs derived from, among others, chemical structure. This could lead to refined chemical designs in early drug discovery. Machine-learning models have many advantages, such as the capacity to automate learning (increasing the speed and scalability of predictions), improved generalizability by learning from multicase historical data, and highlighting statistical and potentially clinically significant relationships between input variables. In contrast, the routinely used mechanistic models (physiologically-based pharmacokinetic models and population pharmacokinetics) are currently considered more interpretable, reliable, and require a smaller sample size of data, although insights differ on a case-by-case basis. Therefore, they may be appropriate for later stages of drug-drug interaction assessment when more in vivo and clinical data are available. A combined approach of using mechanistic models to highlight features that can be used for training machine-learning models may also be exploitable in the future to improve the performance of machine learning. In this review, we provide concepts, strategic considerations, and compare machine learning to mechanistic modeling for drug-drug interaction risk assessment across the stages of drug discovery and development.

Cell Morphological Profiling Enables High-Throughput Screening for PROteolysis TArgeting Chimera (PROTAC) Phenotypic Signature.

PROteolysis TArgeting Chimeras (PROTACs) use the ubiquitin-proteasome system to degrade a protein of interest for therapeutic benefit. Advances made in targeted protein degradation technology have been remarkable, with several molecules having moved into clinical studies. However, robust routes to assess and better understand the safety risks of PROTACs need to be identified, which is an essential step toward delivering efficacious and safe compounds to patients. In this work, we used Cell Painting, an unbiased high-content imaging method, to identify phenotypic signatures of PROTACs. Chemical clustering and model prediction allowed the identification of a mitotoxicity signature that could not be expected by screening the individual PROTAC components. The data highlighted the benefit of unbiased phenotypic methods for identifying toxic signatures and the potential to impact drug design.

Multi-task convolutional neural networks for predicting in vitro clearance endpoints from molecular images.

Optimization of compound metabolic stability is a highly topical issue in pharmaceutical research. Accordingly, application of predictive in silico models can potentially reduce the number of design-make-test-analyze iterations and consequently speed up the progression of novel candidate molecules. Herein, we have investigated the question if multiple in vitro clearance endpoints could be accurately predicted from image-based molecular representations. Thus, compound measurements for four commonly investigated clearance endpoints were curated from AstraZeneca internal sources, providing a sound basis for building multi-task convolutional neural network models. Application of several increasingly challenging data splitting strategies confirmed that convolutional neural network models were successful at capturing implicit chemical relationships contained in training and test data, similar to what is commonly observed for structural fingerprints. Furthermore, model benchmarking against state-of-the-art machine learning methods, including deep neural networks and graph convolutional neural networks, trained with structure- and graph-based representations, respectively, revealed on par or increased accuracy of convolutional neural networks with clear benefit of multi-task learning across all clearance endpoints. Our findings indicate that image-based molecular representations can be applied to predict multiple clearance endpoints, suggesting a potential follow-up to investigate model interpretability from molecular images.

Machine Learning in Chemoinformatics and Medicinal Chemistry.

In chemoinformatics and medicinal chemistry, machine learning has evolved into an important approach. In recent years, increasing computational resources and new deep learning algorithms have put machine learning onto a new level, addressing previously unmet challenges in pharmaceutical research. In silico approaches for compound activity predictions, de novo design, and reaction modeling have been further advanced by new algorithmic developments and the emergence of big data in the field. Herein, novel applications of machine learning and deep learning in chemoinformatics and medicinal chemistry are reviewed. Opportunities and challenges for new methods and applications are discussed, placing emphasis on proper baseline comparisons, robust validation methodologies, and new applicability domains.

Prediction of In Vivo Pharmacokinetic Parameters and Time-Exposure Curves in Rats Using Machine Learning from the Chemical Structure.

Animal pharmacokinetic (PK) data as well as human and animal in vitro systems are utilized in drug discovery to define the rate and route of drug elimination. Accurate prediction and mechanistic understanding of drug clearance and disposition in animals provide a degree of confidence for extrapolation to humans. In addition, prediction of in vivo properties can be used to improve design during drug discovery, help select compounds with better properties, and reduce the number of in vivo experiments. In this study, we generated machine learning models able to predict rat in vivo PK parameters and concentration-time PK profiles based on the molecular chemical structure and either measured or predicted in vitro parameters. The models were trained on internal in vivo rat PK data for over 3000 diverse compounds from multiple projects and therapeutic areas, and the predicted endpoints include clearance and oral bioavailability. We compared the performance of various traditional machine learning algorithms and deep learning approaches, including graph convolutional neural networks. The best models for PK parameters achieved R2 = 0.63 [root mean squared error (RMSE) = 0.26] for clearance and R2 = 0.55 (RMSE = 0.46) for bioavailability. The models provide a fast and cost-efficient way to guide the design of molecules with optimal PK profiles, to enable the prediction of virtual compounds at the point of design, and to drive prioritization of compounds for in vivo assays.

Approach for the Design of Covalent Protein Kinase Inhibitors via Focused Deep Generative Modeling.

Deep machine learning is expanding the conceptual framework and capacity of computational compound design, enabling new applications through generative modeling. We have explored the systematic design of covalent protein kinase inhibitors by learning from kinome-relevant chemical space, followed by focusing on an exemplary kinase of interest. Covalent inhibitors experience a renaissance in drug discovery, especially for targeting protein kinases. However, computational design of this class of inhibitors has thus far only been little investigated. To this end, we have devised a computational approach combining fragment-based design and deep generative modeling augmented by three-dimensional pharmacophore screening. This approach is thought to be particularly relevant for medicinal chemistry applications because it combines knowledge-based elements with deep learning and is chemically intuitive. As an exemplary application, we report for Bruton's tyrosine kinase (BTK), a major drug target for the treatment of inflammatory diseases and leukemia, the generation of novel candidate inhibitors with a specific chemically reactive group for covalent modification, requiring only little target-specific compound information to guide the design efforts. Newly generated compounds include known inhibitors and characteristic substructures and many novel candidates, thus lending credence to the computational approach, which is readily applicable to other targets.

Structure- and Similarity-Based Survey of Allosteric Kinase Inhibitors, Activators, and Closely Related Compounds.

Allosteric kinase inhibitors are thought to have high selectivity and are prime candidates for kinase drug discovery. In addition, the exploration of allosteric mechanisms represents an attractive topic for basic research and drug design. Although the identification and characterization of allosteric kinase inhibitors is still far from being routine, X-ray structures of kinase complexes have been determined for a significant number of such inhibitors. On the basis of structural data, allosteric inhibitors can be confirmed. We report a comprehensive survey of allosteric kinase inhibitors and activators from publicly available X-ray structures, map their binding sites, and determine their distribution over binding pockets in kinases. In addition, we discuss structural features of these compounds and identify active structural analogues and high-confidence target annotations, indicating additional activities for a subset of allosteric inhibitors. This contribution aims to provide a detailed structure-based view of allosteric kinase inhibition.

Impact of Artificial Intelligence on Compound Discovery, Design, and Synthesis.

As in other areas, artificial intelligence (AI) is heavily promoted in different scientific fields, including chemistry. Although chemistry traditionally tends to be a conservative field and slower than others to adapt new concepts, AI is increasingly being investigated across chemical disciplines. In medicinal chemistry, supported by computer-aided drug design and cheminformatics, computational methods have long been employed to aid in the search for and optimization of active compounds. We are currently witnessing a multitude of AI-related publications in the medicinal-chemistry-relevant literature and anticipate that the numbers will further increase. Often, advances through AI promoted in such reports are difficult to reconcile or remain questionable, which hampers the acceptance of computational work in interdisciplinary environments. Herein we attempt to highlight selected investigations in which AI has shown promise to impact medicinal chemistry in areas such as compound design and synthesis.

Machine Learning Models for Human In Vivo Pharmacokinetic Parameters with In-House Validation.

Prior to clinical development, a comprehensive pharmacokinetic characterization of a novel drug is required to understand its exposure at the site of action and elimination. Accordingly, in vitro assays and animal pharmacokinetic studies are regularly employed to predict drug exposure in humans, which is often costly and time-consuming. For this reason, the prediction of human pharmacokinetics at the point of design would be of high value for drug discovery. Therefore, we have established a comprehensive data curation protocol that enables machine learning evaluation of 12 human in vivo pharmacokinetic parameters using only chemical structure information and available doses for 1001 unique compounds. These machine learning models were thoroughly investigated and validated using both an independent hold-out test set and AstraZeneca clinical data. In addition, the availability of preclinical predictions for a subset of internal clinical candidates allowed us to compare our in silico approach with state-of-the-art pharmacokinetic predictions. Based on this evaluation, three fit-for-purpose models for AUC PO (Rtest2 = 0.63; RMSEtest = 0.76), Cmax PO (Rtest2 = 0.68; RMSEtest = 0.62), and Vdss IV (Rtest2 = 0.47; RMSEtest = 0.50) were identified. Based on the findings, our machine learning models have considerable potential for practical applications in drug discovery, such as influencing decision-making in drug discovery projects and progression of drug candidates toward the clinic.

Systematic comparison of competitive and allosteric kinase inhibitors reveals common structural characteristics.

Allosteric and ATP-competitive kinase inhibitors act by distinct mechanisms and are expected to have high and low kinase selectivity, respectively. This also raises the question whether or not these different types of inhibitors might be structurally distinct. To address this question, we have assembled data sets of currently available competitive and allosteric kinase inhibitors confirmed by X-ray crystallography and systematically compared these compounds on the basis of different structural criteria. Many competitive and allosteric inhibitors were found to contain the same or similar substructures and a subset of allosteric inhibitors was found to share core structures with ATP site-directed inhibitors. In some instances, small chemical modifications of common cores were found to yield either allosteric or competitive inhibitors. Hence, these different categories of inhibitors with distinct mechanisms of action were often structurally related and represented much more of a structural continuum than discrete states. Additional target annotations were frequently identified for competitive inhibitors, but were rare for allosteric inhibitors. As a part of this study, our collection of kinase inhibitors and the associated information are made freely available to enable further assessment of chemical modifications that distinguish similar kinase inhibitors with distinct mechanisms of action.

Data set of competitive and allosteric protein kinase inhibitors confirmed by X-ray crystallography.

A data set was generated comprising currently available competitive and allosteric human protein kinase inhibitors confirmed by X-ray crystallography. This data set has been used to systematically explore structural relationships between these types of inhibitors with different mechanisms of action. A major finding of this study has been that these different inhibitor types frequently displayed structural relationships and essentially represented a structural continuum [1]. Use of the data set is not limited to the inhibitor-centric exploration of structural relationships. The collection of kinase inhibitors with structurally confirmed distinct mechanisms of action can also be used, for example, to aid in structure-based drug design or the search for new allosteric kinase inhibitors.

Assessing the information content of structural and protein-ligand interaction representations for the classification of kinase inhibitor binding modes via machine learning and active learning.

For kinase inhibitors, X-ray crystallography has revealed different types of binding modes. Currently, more than 2000 kinase inhibitors with known binding modes are available, which makes it possible to derive and test machine learning models for the prediction of inhibitors with different binding modes. We have addressed this prediction task to evaluate and compare the information content of distinct molecular representations including protein-ligand interaction fingerprints (IFPs) and compound structure-based structural fingerprints (i.e., atom environment/fragment fingerprints). IFPs were designed to capture binding mode-specific interaction patterns at different resolution levels. Accurate predictions of kinase inhibitor binding modes were achieved with random forests using both representations. The performance of IFPs was consistently superior to atom environment fingerprints, albeit only by less than 10%. An active learning strategy applying information entropy-based selection of training instances was applied as a diagnostic approach to assess the relative information content of distinct representations. IFPs were found to capture more binding mode-relevant information than atom environment fingerprints, leading to highly predictive models even when training instances were randomly selected. By contrast, for atom environment fingerprints, the derivation of accurate models via active learning depended on entropy-based selection of informative training compounds. Notably, higher information content of IFPs confirmed by active learning only resulted in small improvements in global prediction accuracy compared to models derived using atom environment fingerprints. For practical applications, prediction of binding modes of new kinase inhibitors on the basis of chemical structure is highly attractive.

Machine Learning Models for Accurate Prediction of Kinase Inhibitors with Different Binding Modes.

Noncovalent inhibitors of protein kinases have different modes of action. They bind to the active or inactive form of kinases, compete with ATP, stabilize inactive kinase conformations, or act through allosteric sites. Accordingly, kinase inhibitors have been classified on the basis of different binding modes. For medicinal chemistry, it would be very useful to derive mechanistic hypotheses for newly discovered inhibitors. Therefore, we have applied different machine learning approaches to generate models for predicting different classes of kinase inhibitors including types I, I1/2, and II as well as allosteric inhibitors. These models were built on the basis of compounds with binding modes confirmed by X-ray crystallography and yielded unexpectedly accurate and stable predictions without the need for deep learning. The results indicate that the new machine learning models have considerable potential for practical applications. Therefore, our data sets and models are made freely available.

Data structures for computational compound promiscuity analysis and exemplary applications to inhibitors of the human kinome.

Small molecules with multi-target activity, also termed promiscuous compounds, are increasingly considered for pharmaceutical applications. The use of promiscuous chemical entities represents a departure from the compound specificity paradigm, one of the pillars of modern drug discovery. The popularity of promiscuous compounds is due to the concept of polypharmacology; another more recent drug discovery paradigm. It refers to insights that the efficacy of drugs often depends on interactions with multiple targets. Views concerning the extent to which small molecules might form well-defined interactions with multiple targets often differ, but comprehensive experimental investigations of promiscuity are currently rare. On the other hand, large volumes of active compounds and experimental measurements are becoming available and enable data-driven analyses of compound selectivity versus promiscuity. In this perspective, we discuss computational methods and data structures designed for promiscuity analysis. In addition, findings from large-scale exploration of activity profiles of inhibitors covering the human kinome are summarized. Although many kinase inhibitors are expected to be promiscuous, they are frequently found to be selective, which provides opportunities for target-directed drug discovery (rather than polypharmacology). We also discuss that machine learning yields evidence for the existence of structure-promiscuity relationships.

Promiscuity analysis of a kinase panel screen with designated p38 alpha inhibitors.

Protein phosphorylation by kinases is of critical importance for the regulation of many cellular functions. When kinases are deregulated numerous biological processes are affected, which may cause a variety of diseases. Therefore, kinase inhibition plays an important role for therapeutic intervention. A number of kinase inhibitors have been approved as drugs, initially in oncology where promiscuous (multi-kinase) inhibitors were most efficacious. Exploring kinase inhibitor selectivity and promiscuity for therapy is among the most challenging aspects of kinase drug discovery. Herein, we thoroughly analyze a kinase profiling experiment in which 637 designated inhibitors of p38α MAP kinase (p38α) were tested against a panel of 60 kinases distributed across the human kinome. In this experiment, only 19% of the inhibitors were found to be promiscuous when the median p38α inhibition level was applied as an activity threshold. Promiscuous inhibitors had a median value of two targets per compound, and many of these inhibitors were only active against the p38α and closely related JNK3 enzymes. Promiscuity cliffs were identified and analyzed in a network representation revealing structural modifications that were implicated in triggering compound promiscuity. Taken together, the findings revealed a high degree of selectivity of designated p38α directed inhibitors although they target the ATP binding site that is largely conserved across the human kinome.

Identifying Promiscuous Compounds with Activity against Different Target Classes.

Compounds with multitarget activity are of high interest for polypharmacological drug discovery. Such promiscuous compounds might be active against closely related target proteins from the same family or against distantly related or unrelated targets. Compounds with activity against distinct targets are not only of interest for polypharmacology but also to better understand how small molecules might form specific interactions in different binding site environments. We have aimed to identify compounds with activity against drug targets from different classes. To these ends, a systematic analysis of public biological screening data was carried out. Care was taken to exclude compounds from further consideration that were prone to experimental artifacts and false positive activity readouts. Extensively assayed compounds were identified and found to contain molecules that were consistently inactive in all assays, active against a single target, or promiscuous. The latter included more than 1000 compounds that were active against 10 or more targets from different classes. These multiclass ligands were further analyzed and exemplary compounds were found in X-ray structures of complexes with distinct targets. Our collection of multiclass ligands should be of interest for pharmaceutical applications and further exploration of binding characteristics at the molecular level. Therefore, these highly promiscuous compounds are made publicly available.

Data structures for compound promiscuity analysis: promiscuity cliffs, pathways and promiscuity hubs formed by inhibitors of the human kinome.

AIM: A large collection of promiscuity cliffs (PCs), PC pathways (PCPs) and promiscuity hubs (PHs) formed by inhibitors of human kinases is made freely available. METHODOLOGY: Inhibitor PCs were systematically identified and organized in network representations, from which PCPs were extracted. PH compounds were classified and their neighborhoods analyzed. DATA & EXEMPLARY RESULTS: Nearly 16,000 PCs covering the human kinome were identified, which yielded more than 600 PC clusters and 8900 PCPs. Moreover, 520 PHs were obtained. LIMITATIONS & NEXT STEPS: PC and PCP data structures capture structure-promiscuity relationships. Promiscuity assessment is also affected by data sparseness. Given the rapid growth of kinase inhibitor data, the relevance of PC/PCP/PH information for medicinal chemistry and chemical biology applications will further increase.