Unprecedently Large-Scale Kinase Inhibitor Set Enabling the Accurate Prediction of Compound-Kinase Activities: A Way toward Selective Promiscuity by Design?

Poor Generalization by Current Deep Learning Models for Predicting Binding Affinities of Kinase Inhibitors

The extreme surge of interest over the past decade surrounding the use of neural networks has inspired many groups to deploy them for predicting binding affinities of drug-like molecules to their receptors. A model that can accurately make such predictions has the potential to screen large chemical libraries and help streamline the drug discovery process. However, despite reports of models that accurately predict quantitative inhibition using protein kinase sequences and inhibitors' SMILES strings, it is still unclear whether these models can generalize to previously unseen data. Here, we build a Convolutional Neural Network (CNN) analogous to those previously reported and evaluate the model over four datasets commonly used for inhibitor/kinase predictions. We find that the model performs comparably to those previously reported, provided that the individual data points are randomly split between the training set and the test set. However, model performance is dramatically deteriorated when all data for a given inhibitor is placed together in the same training/testing fold, implying that information leakage underlies the models' performance. Through comparison to simple models in which the SMILES strings are tokenized, or in which test set predictions are simply copied from the closest training set data points, we demonstrate that there is essentially no generalization whatsoever in this model. In other words, the model has not learned anything about molecular interactions, and does not provide any benefit over much simpler and more transparent models. These observations strongly point to the need for richer structure-based encodings, to obtain useful prospective predictions of not-yet-synthesized candidate inhibitors.

Large-Scale Modeling of Sparse Protein Kinase Activity Data

Protein kinases are a protein family that plays an important role in several complex diseases such as cancer and cardiovascular and immunological diseases. Protein kinases have conserved ATP binding sites, which when targeted can lead to similar activities of inhibitors against different kinases. This can be exploited to create multitarget drugs. On the other hand, selectivity (lack of similar activities) is desirable in order to avoid toxicity issues. There is a vast amount of protein kinase activity data in the public domain, which can be used in many different ways. Multitask machine learning models are expected to excel for these kinds of data sets because they can learn from implicit correlations between tasks (in this case activities against a variety of kinases). However, multitask modeling of sparse data poses two major challenges: (i) creating a balanced train-test split without data leakage and (ii) handling missing data. In this work, we construct a protein kinase benchmark set composed of two balanced splits without data leakage, using random and dissimilarity-driven cluster-based mechanisms, respectively. This data set can be used for benchmarking and developing protein kinase activity prediction models. Overall, the performance on the dissimilarity-driven cluster-based split is lower than on random split-based sets for all models, indicating poor generalizability of models. Nevertheless, we show that multitask deep learning models, on this very sparse data set, outperform single-task deep learning and tree-based models. Finally, we demonstrate that data imputation does not improve the performance of (multitask) models on this benchmark set.

Establishment of extensive artificial intelligence models for kinase inhibitor prediction: Identification of novel PDGFRB inhibitors

Identifying hit compounds is an important step in drug development. Unfortunately, this process continues to be a challenging task. Several machine learning models have been generated to aid in simplifying and improving the prediction of candidate compounds. Models tuned for predicting kinase inhibitors have been established. However, an effective model can be limited by the size of the chosen training dataset. In this study, we tested several machine learning models to predict potential kinase inhibitors. A dataset was curated from a number of publicly available repositories. This resulted in a comprehensive dataset covering more than half of the human kinome. More than 2,000 kinase models were established using different model approaches. The performances of the models were compared, and the Keras-MLP model was determined to be the best performing model. The model was then used to screen a chemical library for potential inhibitors targeting platelet-derived growth factor receptor-β (PDGFRB). Several PDGFRB candidates were selected, and in vitro assays confirmed four compounds with PDGFRB inhibitory activity and IC₅₀ values in the nanomolar range. These results show the effectiveness of machine learning models trained on the reported dataset. This report would aid in the establishment of machine learning models as well as in the discovery of novel kinase inhibitors.

Proteochemometric Modeling Identifies Chemically Diverse Norepinephrine Transporter Inhibitors

Solute carriers (SLCs) are relatively underexplored compared to other prominent protein families such as kinases and G protein-coupled receptors. However, proteins from the SLC family play an essential role in various diseases. One such SLC is the high-affinity norepinephrine transporter (NET/SLC6A2). In contrast to most other SLCs, the NET has been relatively well studied. However, the chemical space of known ligands has a low chemical diversity, making it challenging to identify chemically novel ligands. Here, a computational screening pipeline was developed to find new NET inhibitors. The approach increases the chemical space to model for NETs using the chemical space of related proteins that were selected utilizing similarity networks. Prior proteochemometric models added data from related proteins, but here we use a data-driven approach to select the optimal proteins to add to the modeled data set. After optimizing the data set, the proteochemometric model was optimized using stepwise feature selection. The final model was created using a two-step approach combining several proteochemometric machine learning models through stacking. This model was applied to the extensive virtual compound database of Enamine, from which the top predicted 22,000 of the 600 million virtual compounds were clustered to end up with 46 chemically diverse candidates. A subselection of 32 candidates was synthesized and subsequently tested using an impedance-based assay. There were five hit compounds identified (hit rate 16%) with sub-micromolar inhibitory potencies toward NET, which are promising for follow-up experimental research. This study demonstrates a data-driven approach to diversify known chemical space to identify novel ligands and is to our knowledge the first to select this set based on the sequence similarity of related targets.

Papyrus: a large-scale curated dataset aimed at bioactivity predictions

With the ongoing rapid growth of publicly available ligand-protein bioactivity data, there is a trove of valuable data that can be used to train a plethora of machine-learning algorithms. However, not all data is equal in terms of size and quality and a significant portion of researchers' time is needed to adapt the data to their needs. On top of that, finding the right data for a research question can often be a challenge on its own. To meet these challenges, we have constructed the Papyrus dataset. Papyrus is comprised of around 60 million data points. This dataset contains multiple large publicly available datasets such as ChEMBL and ExCAPE-DB combined with several smaller datasets containing high-quality data. The aggregated data has been standardised and normalised in a manner that is suitable for machine learning. We show how data can be filtered in a variety of ways and also perform some examples of quantitative structure-activity relationship analyses and proteochemometric modelling. Our ambition is that this pruned data collection constitutes a benchmark set that can be used for constructing predictive models, while also providing an accessible data source for research.

KUALA: a machine learning-driven framework for kinase inhibitors repositioning

The family of protein kinases comprises more than 500 genes involved in numerous functions. Hence, their physiological dysfunction has paved the way toward drug discovery for cancer, cardiovascular, and inflammatory diseases. As a matter of fact, Kinase binding sites high similarity has a double role. On the one hand it is a critical issue for selectivity, on the other hand, according to poly-pharmacology, a synergistic controlled effect on more than one target could be of great pharmacological interest. Another important aspect of binding similarity is the possibility of exploit it for repositioning of drugs on targets of the same family. In this study, we propose our approach called Kinase drUgs mAchine Learning frAmework (KUALA) to automatically identify kinase active ligands by using specific sets of molecular descriptors and provide a multi-target priority score and a repurposing threshold to suggest the best repurposable and non-repurposable molecules. The comprehensive list of all kinase-ligand pairs and their scores can be found at https://github.com/molinfrimed/multi-kinases .

Sequence-based prediction of protein binding regions and drug-target interactions

Identifying drug-target interactions (DTIs) is important for drug discovery. However, searching all drug-target spaces poses a major bottleneck. Therefore, recently many deep learning models have been proposed to address this problem. However, the developers of these deep learning models have neglected interpretability in model construction, which is closely related to a model's performance. We hypothesized that training a model to predict important regions on a protein sequence would increase DTI prediction performance and provide a more interpretable model. Consequently, we constructed a deep learning model, named Highlights on Target Sequences (HoTS), which predicts binding regions (BRs) between a protein sequence and a drug ligand, as well as DTIs between them. To train the model, we collected complexes of protein-ligand interactions and protein sequences of binding sites and pretrained the model to predict BRs for a given protein sequence-ligand pair via object detection employing transformers. After pretraining the BR prediction, we trained the model to predict DTIs from a compound token designed to assign attention to BRs. We confirmed that training the BRs prediction model indeed improved the DTI prediction performance. The proposed HoTS model showed good performance in BR prediction on independent test datasets even though it does not use 3D structure information in its prediction. Furthermore, the HoTS model achieved the best performance in DTI prediction on test datasets. Additional analysis confirmed the appropriate attention for BRs and the importance of transformers in BR and DTI prediction. The source code is available on GitHub ( https://github.com/GIST-CSBL/HoTS ).

A human-based multi-gene signature enables quantitative drug repurposing for metabolic disease

Insulin resistance (IR) contributes to the pathophysiology of diabetes, dementia, viral infection, and cardiovascular disease. Drug repurposing (DR) may identify treatments for IR; however, barriers include uncertainty whether in vitro transcriptomic assays yield quantitative pharmacological data, or how to optimise assay design to best reflect in vivo human disease. We developed a clinical-based human tissue IR signature by combining lifestyle-mediated treatment responses (>500 human adipose and muscle biopsies) with biomarkers of disease status (fasting IR from >1200 biopsies). The assay identified a chemically diverse set of >130 positively acting compounds, highly enriched in true positives, that targeted 73 proteins regulating IR pathways. Our multi-gene RNA assay score reflected the quantitative pharmacological properties of a set of epidermal growth factor receptor-related tyrosine kinase inhibitors, providing insight into drug target specificity; an observation supported by deep learning-based genome-wide predicted pharmacology. Several drugs identified are suitable for evaluation in patients, particularly those with either acute or severe chronic IR.

Exploration and augmentation of pharmacological space via adversarial auto-encoder model for facilitating kinase-centric drug development

Predicting compound-protein interactions (CPIs) is of great importance for drug discovery and repositioning, yet still challenging mainly due to the sparse nature of CPI matrixes, resulting in poor generalization performance. Hence, unlike typical CPI prediction models focused on representation learning or model selection, we propose a deep neural network-based strategy, PCM-AAE, that re-explores and augments the pharmacological space of kinase inhibitors by introducing the adversarial auto-encoder model (AAE) to improve the generalization of the prediction model. To complete the data space, we constructed Ensemble of PCM-AAE (EPA), an ensemble model that quickly and accurately yields quantitative predictions of binding affinity between any human kinase and inhibitor. In rigorous internal validation, EPA showed excellent performance, consistently outperforming the model trained with the imbalanced set, especially for targets with relatively fewer training data points. Improved prediction accuracy of EPA for external datasets enhances its generalization ability, making it possible to gracefully handle previously unseen kinases and inhibitors. EPA showed promising potential when directly applied to virtual screening and off-target prediction, exhibiting its practicality in hit prediction. Our strategy is expected to facilitate kinase-centric drug development, as well as to solve more challenging prediction problems with insufficient data points.

Facing small and biased data dilemma in drug discovery with enhanced federated learning approaches

Artificial intelligence (AI) models usually require large amounts of high-quality training data, which is in striking contrast to the situation of small and biased data faced by current drug discovery pipelines. The concept of federated learning has been proposed to utilize distributed data from different sources without leaking sensitive information of the data. This emerging decentralized machine learning paradigm is expected to dramatically improve the success rate of AI-powered drug discovery. Here, we simulated the federated learning process with different property and activity datasets from different sources, among which overlapping molecules with high or low biases exist in the recorded values. Beyond the benefit of gaining more data, we also demonstrated that federated training has a regularization effect superior to centralized training on the pooled datasets with high biases. Moreover, different network architectures for clients and aggregation algorithms for coordinators have been compared on the performance of federated learning, where personalized federated learning shows promising results. Our work demonstrates the applicability of federated learning in predicting drug-related properties and highlights its promising role in addressing the small and biased data dilemma in drug discovery.

Type II Binders Targeting the "GLR-Out" Conformation of the Pseudokinase STRADα

Pseudokinases play important roles in signal transduction and cellular processes similar to those of catalytically competent kinases. However, pseudokinase pharmacological tractability and conformational space accessibility are poorly understood. Pseudokinases have only recently been suggested to adopt "inactive" conformations or interact with conformation-specific kinase inhibitors (e.g., type II compounds). In this work, the heavily substituted pseudokinase STRADα, which possesses a DFG → GLR substitution in the catalytic site that permits nucleotide binding while impairing divalent cation coordination, is used as a test case to demonstrate the potential applicability of conformation-specific, type II compounds to pseudokinase pharmacology. Integrated structural modeling is employed to generate a "GLR-out" conformational ensemble. Likely interacting type II compounds are identified through virtual screening against this ensemble model. Biophysical validation of compound binding is demonstrated through protein thermal stabilization and ATP competition. Localization of a top-performing compound through surface methylation strongly suggests that STRADα can adopt the "GLR-out" conformation and interact with compounds that comply with the standard type II pharmacophore. These results suggest that, despite a loss of catalytic function, some pseudokinases, including STRADα, may retain the conformational switching properties of conventional protein kinases.

The Kinase Chemogenomic Set (KCGS): An Open Science Resource for Kinase Vulnerability Identification

We describe the assembly and annotation of a chemogenomic set of protein kinase inhibitors as an open science resource for studying kinase biology. The set only includes inhibitors that show potent kinase inhibition and a narrow spectrum of activity when screened across a large panel of kinase biochemical assays. Currently, the set contains 187 inhibitors that cover 215 human kinases. The kinase chemogenomic set (KCGS), current Version 1.0, is the most highly annotated set of selective kinase inhibitors available to researchers for use in cell-based screens.

Is Structure-Based Drug Design Ready for Selectivity Optimization?

Alchemical free-energy calculations are now widely used to drive or maintain potency in small-molecule lead optimization with a roughly 1 kcal/mol accuracy. Despite this, the potential to use free-energy calculations to drive optimization of compound selectivity among two similar targets has been relatively unexplored in published studies. In the most optimistic scenario, the similarity of binding sites might lead to a fortuitous cancellation of errors and allow selectivity to be predicted more accurately than affinity. Here, we assess the accuracy with which selectivity can be predicted in the context of small-molecule kinase inhibitors, considering the very similar binding sites of human kinases CDK2 and CDK9 as well as another series of ligands attempting to achieve selectivity between the more distantly related kinases CDK2 and ERK2. Using a Bayesian analysis approach, we separate systematic from statistical errors and quantify the correlation in systematic errors between selectivity targets. We find that, in the CDK2/CDK9 case, a high correlation in systematic errors suggests that free-energy calculations can have significant impact in aiding chemists in achieving selectivity, while in more distantly related kinases (CDK2/ERK2), the correlation in systematic error suggests that fortuitous cancellation may even occur between systems that are not as closely related. In both cases, the correlation in systematic error suggests that longer simulations are beneficial to properly balance statistical error with systematic error to take full advantage of the increase in apparent free-energy calculation accuracy in selectivity prediction.

Developing a Kinase-Specific Target Selection Method Using a Structure-Based Machine Learning Approach

Introduction

Despite recent advances in the drug discovery field, developing selective kinase inhibitors remains a complicated issue for a number of reasons, one of which is that there are striking structural similarities in the ATP-binding pockets of kinases.

Objective

To address this problem, we have designed a machine learning model utilizing various structure-based and energy-based descriptors to better characterize protein–ligand interactions.

Methods

In this work, we use a dataset of 104 human kinases with available PDB structures and experimental activity data against 1202 small-molecule compounds from the PubChem BioAssay dataset “Navigating the Kinome”. We propose structure-based interaction descriptors to build activity predicting machine learning model.

Results and Discussion

We report a ligand-oriented computational method for accurate kinase target prioritizing. Our method shows high accuracy compared to similar structure-based activity prediction methods, and more importantly shows the same prediction accuracy when tested on the special set of structurally remote compounds, showing that it is unbiased to ligand structural similarity in the training set data. We hope that our approach will be useful for the development of novel highly selective kinase inhibitors.

Ibrutinib-Mediated Atrial Fibrillation Attributable to Inhibition of C-Terminal Src Kinase

BACKGROUND

Ibrutinib is a Bruton tyrosine kinase inhibitor with remarkable efficacy against B-cell cancers. Ibrutinib also increases the risk of atrial fibrillation (AF), which remains poorly understood.

METHODS

We performed electrophysiology studies on mice treated with ibrutinib to assess inducibility of AF. Chemoproteomic analysis of cardiac lysates identified candidate ibrutinib targets, which were further evaluated in genetic mouse models and additional pharmacological experiments. The pharmacovigilance database, VigiBase, was queried to determine whether drug inhibition of an identified candidate kinase was associated with increased reporting of AF.

RESULTS

We demonstrate that treatment of mice with ibrutinib for 4 weeks results in inducible AF, left atrial enlargement, myocardial fibrosis, and inflammation. This effect was reproduced in mice lacking Bruton tyrosine kinase, but not in mice treated with 4 weeks of acalabrutinib, a more specific Bruton tyrosine kinase inhibitor, demonstrating that AF is an off-target side effect. Chemoproteomic profiling identified a short list of candidate kinases that was narrowed by additional experimentation leaving CSK (C-terminal Src kinase) as the strongest candidate for ibrutinib-induced AF. Cardiac-specific Csk knockout in mice led to increased AF, left atrial enlargement, fibrosis, and inflammation, phenocopying ibrutinib treatment. Disproportionality analyses in VigiBase confirmed increased reporting of AF associated with kinase inhibitors blocking Csk versus non-Csk inhibitors, with a reporting odds ratio of 8.0 (95% CI, 7.3-8.7; P<0.0001).

CONCLUSIONS

These data identify Csk inhibition as the mechanism through which ibrutinib leads to AF. Registration: URL: https://ww.clinicaltrials.gov; Unique identifier: NCT03530215.

Successive Statistical and Structure-Based Modeling to Identify Chemically Novel Kinase Inhibitors

Kinases are frequently studied in the context of anticancer drugs. Their involvement in cell responses, such as proliferation, differentiation, and apoptosis, makes them interesting subjects in multitarget drug design. In this study, a workflow is presented that models the bioactivity spectra for two panels of kinases: (1) inhibition of RET, BRAF, SRC, and S6K, while avoiding inhibition of MKNK1, TTK, ERK8, PDK1, and PAK3, and (2) inhibition of AURKA, PAK1, FGFR1, and LKB1, while avoiding inhibition of PAK3, TAK1, and PIK3CA. Both statistical and structure-based models were included, which were thoroughly benchmarked and optimized. A virtual screening was performed to test the workflow for one of the main targets, RET kinase. This resulted in 5 novel and chemically dissimilar RET inhibitors with remaining RET activity of <60% (at a concentration of 10 μM) and similarities with known RET inhibitors from 0.18 to 0.29 (Tanimoto, ECFP6). The four more potent inhibitors were assessed in a concentration range and proved to be modestly active with a pIC₅₀ value of 5.1 for the most active compound. The experimental validation of inhibitors for RET strongly indicates that the multitarget workflow is able to detect novel inhibitors for kinases, and hence, this workflow can potentially be applied in polypharmacology modeling. We conclude that this approach can identify new chemical matter for existing targets. Moreover, this workflow can easily be applied to other targets as well.

Practical Applications of Deep Learning To Impute Heterogeneous Drug Discovery Data

Contemporary deep learning approaches still struggle to bring a useful improvement in the field of drug discovery because of the challenges of sparse, noisy, and heterogeneous data that are typically encountered in this context. We use a state-of-the-art deep learning method, Alchemite, to impute data from drug discovery projects, including multitarget biochemical activities, phenotypic activities in cell-based assays, and a variety of absorption, distribution, metabolism, and excretion (ADME) endpoints. The resulting model gives excellent predictions for activity and ADME endpoints, offering an average increase in R² of 0.22 versus quantitative structure-activity relationship methods. The model accuracy is robust to combining data across uncorrelated endpoints and projects with different chemical spaces, enabling a single model to be trained for all compounds and endpoints. We demonstrate improvements in accuracy on the latest chemistry and data when updating models with new data as an ongoing medicinal chemistry project progresses.

In silico Methods for Design of Kinase Inhibitors as Anticancer Drugs

Rational drug design implies usage of molecular modeling techniques such as pharmacophore modeling, molecular dynamics, virtual screening, and molecular docking to explain the activity of biomolecules, define molecular determinants for interaction with the drug target, and design more efficient drug candidates. Kinases play an essential role in cell function and therefore are extensively studied targets in drug design and discovery. Kinase inhibitors are clinically very important and widely used antineoplastic drugs. In this review, computational methods used in rational drug design of kinase inhibitors are discussed and compared, considering some representative case studies.

Proteochemometrics - recent developments in bioactivity and selectivity modeling

Proteochemometrics is a machine learning based modeling approach relying on a combination of ligand and protein descriptors. With ongoing developments in machine learning and increases in public data the technique is more frequently applied in early drug discovery, typically in ligand-target binding prediction. Common applications include improvements to single target quantitative structure-activity relationship models, protein selectivity and promiscuity modeling, and large-scale deep learning approaches. The increase in predictive power using proteochemometrics is observed in multi-target bioactivity modeling, opening the door to more extensive studies covering whole protein families. On top of that, with deep learning fueling more complex and larger scale models, proteochemometrics allows faster and higher quality computational models supporting the design, make, test cycle.

Rational discovery of dual-indication multi-target PDE/Kinase inhibitor for precision anti-cancer therapy using structural systems pharmacology

Many complex diseases such as cancer are associated with multiple pathological manifestations. Moreover, the therapeutics for their treatments often lead to serious side effects. Thus, it is needed to develop multi-indication therapeutics that can simultaneously target multiple clinical indications of interest and mitigate the side effects. However, conventional one-drug-one-gene drug discovery paradigm and emerging polypharmacology approach rarely tackle the challenge of multi-indication drug design. For the first time, we propose a one-drug-multi-target-multi-indication strategy. We develop a novel structural systems pharmacology platform 3D-REMAP that uses ligand binding site comparison and protein-ligand docking to augment sparse chemical genomics data for the machine learning model of genome-scale chemical-protein interaction prediction. Experimentally validated predictions systematically show that 3D-REMAP outperforms state-of-the-art ligand-based, receptor-based, and machine learning methods alone. As a proof-of-concept, we utilize the concept of drug repurposing that is enabled by 3D-REMAP to design dual-indication anti-cancer therapy. The repurposed drug can demonstrate anti-cancer activity for cancers that do not have effective treatment as well as reduce the risk of heart failure that is associated with all types of existing anti-cancer therapies. We predict that levosimendan, a PDE inhibitor for heart failure, inhibits serine/threonine-protein kinase RIOK1 and other kinases. Subsequent experiments and systems biology analyses confirm this prediction, and suggest that levosimendan is active against multiple cancers, notably lymphoma, through the direct inhibition of RIOK1 and RNA processing pathway. We further develop machine learning models to predict cancer cell-line's and a patient's response to levosimendan. Our findings suggest that levosimendan can be a promising novel lead compound for the development of safe, effective, and precision multi-indication anti-cancer therapy. This study demonstrates the potential of structural systems pharmacology in designing polypharmacology for precision medicine. It may facilitate transforming the conventional one-drug-one-gene-one-disease drug discovery process and single-indication polypharmacology approach into a new one-drug-multi-target-multi-indication paradigm for complex diseases.

Predictive models of aqueous solubility of organic compounds built on A large dataset of high integrity

Aqueous solubility is one of the most important properties in drug discovery, as it has profound impact on various drug properties, including biological activity, pharmacokinetics (PK), toxicity, and in vivo efficacy. Both kinetic and thermodynamic solubilities are determined during different stages of drug discovery and development. Since kinetic solubility is more relevant in preclinical drug discovery research, especially during the structure optimization process, we have developed predictive models for kinetic solubility with in-house data generated from 11,780 compounds collected from over 200 NCATS intramural research projects. This represents one of the largest kinetic solubility datasets of high quality and integrity. Based on the customized atom type descriptors, the support vector classification (SVC) models were trained on 80% of the whole dataset, and exhibited high predictive performance for estimating the solubility of the remaining 20% compounds within the test set. The values of the area under the receiver operating characteristic curve (AUC-ROC) for the compounds in the test sets reached 0.93 and 0.91, when the threshold for insoluble compounds was set to 10 and 50 μg/mL respectively. The predictive models of aqueous solubility can be used to identify insoluble compounds in drug discovery pipeline, provide design ideas for improving solubility by analyzing the atom types associated with poor solubility and prioritize compound libraries to be purchased or synthesized.

Imputation of Assay Bioactivity Data Using Deep Learning

We describe a novel deep learning neural network method and its application to impute assay pIC₅₀ values. Unlike conventional machine learning approaches, this method is trained on sparse bioactivity data as input, typical of that found in public and commercial databases, enabling it to learn directly from correlations between activities measured in different assays. In two case studies on public domain data sets we show that the neural network method outperforms traditional quantitative structure-activity relationship (QSAR) models and other leading approaches. Furthermore, by focusing on only the most confident predictions the accuracy is increased to R² > 0.9 using our method, as compared to R² = 0.44 when reporting all predictions.

Drug Discovery Maps, a Machine Learning Model That Visualizes and Predicts Kinome-Inhibitor Interaction Landscapes

The interpretation of high-dimensional structure-activity data sets in drug discovery to predict ligand-protein interaction landscapes is a challenging task. Here we present Drug Discovery Maps (DDM), a machine learning model that maps the activity profile of compounds across an entire protein family, as illustrated here for the kinase family. DDM is based on the t-distributed stochastic neighbor embedding (t-SNE) algorithm to generate a visualization of molecular and biological similarity. DDM maps chemical and target space and predicts the activities of novel kinase inhibitors across the kinome. The model was validated using independent data sets and in a prospective experimental setting, where DDM predicted new inhibitors for FMS-like tyrosine kinase 3 (FLT3), a therapeutic target for the treatment of acute myeloid leukemia. Compounds were resynthesized, yielding highly potent, cellularly active FLT3 inhibitors. Biochemical assays confirmed most of the predicted off-targets. DDM is further unique in that it is completely open-source and available as a ready-to-use executable to facilitate broad and easy adoption.

How to Achieve Better Results Using PASS-Based Virtual Screening: Case Study for Kinase Inhibitors

Discovery of new pharmaceutical substances is currently boosted by the possibility of utilization of the Synthetically Accessible Virtual Inventory (SAVI) library, which includes about 283 million molecules, each annotated with a proposed synthetic one-step route from commercially available starting materials. The SAVI database is well-suited for ligand-based methods of virtual screening to select molecules for experimental testing. In this study, we compare the performance of three approaches for the analysis of structure-activity relationships that differ in their criteria for selecting of "active" and "inactive" compounds included in the training sets. PASS (Prediction of Activity Spectra for Substances), which is based on a modified Naïve Bayes algorithm, was applied since it had been shown to be robust and to provide good predictions of many biological activities based on just the structural formula of a compound even if the information in the training set is incomplete. We used different subsets of kinase inhibitors for this case study because many data are currently available on this important class of drug-like molecules. Based on the subsets of kinase inhibitors extracted from the ChEMBL 20 database we performed the PASS training, and then applied the model to ChEMBL 23 compounds not yet present in ChEMBL 20 to identify novel kinase inhibitors. As one may expect, the best prediction accuracy was obtained if only the experimentally confirmed active and inactive compounds for distinct kinases in the training procedure were used. However, for some kinases, reasonable results were obtained even if we used merged training sets, in which we designated as inactives the compounds not tested against the particular kinase. Thus, depending on the availability of data for a particular biological activity, one may choose the first or the second approach for creating ligand-based computational tools to achieve the best possible results in virtual screening.

Clinical pharmacodynamic/exposure characterisation of the multikinase inhibitor ilorasertib (ABT-348) in a phase 1 dose-escalation trial

Background

Ilorasertib (ABT-348) inhibits Aurora and VEGF receptor (VEGFR) kinases. Patients with advanced solid tumours participated in a phase 1 dose-escalation trial to profile the safety, tolerability, and pharmacokinetics of ilorasertib.

Methods

Ilorasertib monotherapy was administered at 10–180 mg orally once daily (Arm I, n = 23), 40–340 mg orally twice daily (Arm II, n = 28), or 8–32 mg intravenously once daily (Arm III, n = 7), on days 1, 8, and 15 of each 28-day cycle.

Results

Dose-limiting toxicities were predominantly related to VEGFR inhibition. The most frequent treatment-emergent adverse events ( > 30%) were: fatigue (48%), anorexia (34%), and hypertension (34%). Pharmacodynamic markers suggested that ilorasertib engaged VEGFR2 and Aurora B kinase, with the VEGFR2 effects reached at lower doses and exposures than Aurora inhibition effects. In Arm II, one basal cell carcinoma patient (40 mg twice daily (BID)) and one patient with adenocarcinoma of unknown primary site (230 mg BID) had partial responses.

Conclusions

In patients with advanced solid tumours, ilorasertib treatment resulted in evidence of engagement of the intended targets and antitumour activity, but with maximum inhibition of VEGFR family kinases occurring at lower exposures than typically required for inhibition of Aurora B in tissue.

Clinical Trial Registration: NCT01110486

The Future of Computational Chemogenomics

Following the elucidation of the human genome, chemogenomics emerged in the beginning of the twenty-first century as an interdisciplinary research field with the aim to accelerate target and drug discovery by making best usage of the genomic data and the data linkable to it. What started as a systematization approach within protein target families now encompasses all types of chemical compounds and gene products. A key objective of chemogenomics is the establishment, extension, analysis, and prediction of a comprehensive SAR matrix which by application will enable further systematization in drug discovery. Herein we outline future perspectives of chemogenomics including the extension to new molecular modalities, or the potential extension beyond the pharma to the agro and nutrition sectors, and the importance for environmental protection. The focus is on computational sciences with potential applications for compound library design, virtual screening, hit assessment, analysis of phenotypic screens, lead finding and optimization, and systems biology-based prediction of toxicology and translational research.

Target and Tissue Selectivity Prediction by Integrated Mechanistic Pharmacokinetic-Target Binding and Quantitative Structure Activity Modeling

Selectivity is an important attribute of effective and safe drugs, and prediction of in vivo target and tissue selectivity would likely improve drug development success rates. However, a lack of understanding of the underlying (pharmacological) mechanisms and availability of directly applicable predictive methods complicates the prediction of selectivity. We explore the value of combining physiologically based pharmacokinetic (PBPK) modeling with quantitative structure-activity relationship (QSAR) modeling to predict the influence of the target dissociation constant (K _D) and the target dissociation rate constant on target and tissue selectivity. The K _D values of CB1 ligands in the ChEMBL database are predicted by QSAR random forest (RF) modeling for the CB1 receptor and known off-targets (TRPV1, mGlu5, 5-HT1a). Of these CB1 ligands, rimonabant, CP-55940, and Δ⁸-tetrahydrocanabinol, one of the active ingredients of cannabis, were selected for simulations of target occupancy for CB1, TRPV1, mGlu5, and 5-HT1a in three brain regions, to illustrate the principles of the combined PBPK-QSAR modeling. Our combined PBPK and target binding modeling demonstrated that the optimal values of the K _D and k _off for target and tissue selectivity were dependent on target concentration and tissue distribution kinetics. Interestingly, if the target concentration is high and the perfusion of the target site is low, the optimal K _D value is often not the lowest K _D value, suggesting that optimization towards high drug-target affinity can decrease the benefit-risk ratio. The presented integrative structure-pharmacokinetic-pharmacodynamic modeling provides an improved understanding of tissue and target selectivity.

Protocols for the Design of Kinase-focused Compound Libraries

Protocols for the design of kinase-focused compound libraries are presented. Kinase-focused compound libraries can be differentiated based on the design goal. Depending on whether the library should be a discovery library specific for one particular kinase, a general discovery library for multiple distinct kinase projects, or even phenotypic screening, there exists today a variety of in silico methods to design candidate compound libraries. We address the following scenarios: 1) Datamining of SAR databases and kinase focused vendor catalogues; 2) Predictions and virtual screening; 3) Structure-based design of combinatorial kinase inhibitors; 4) Design of covalent kinase inhibitors; 5) Design of macrocyclic kinase inhibitors; and 6) Design of allosteric kinase inhibitors and activators.

Beyond the hype: deep neural networks outperform established methods using a ChEMBL bioactivity benchmark set

The increase of publicly available bioactivity data in recent years has fueled and catalyzed research in chemogenomics, data mining, and modeling approaches. As a direct result, over the past few years a multitude of different methods have been reported and evaluated, such as target fishing, nearest neighbor similarity-based methods, and Quantitative Structure Activity Relationship (QSAR)-based protocols. However, such studies are typically conducted on different datasets, using different validation strategies, and different metrics. In this study, different methods were compared using one single standardized dataset obtained from ChEMBL, which is made available to the public, using standardized metrics (BEDROC and Matthews Correlation Coefficient). Specifically, the performance of Naïve Bayes, Random Forests, Support Vector Machines, Logistic Regression, and Deep Neural Networks was assessed using QSAR and proteochemometric (PCM) methods. All methods were validated using both a random split validation and a temporal validation, with the latter being a more realistic benchmark of expected prospective execution. Deep Neural Networks are the top performing classifiers, highlighting the added value of Deep Neural Networks over other more conventional methods. Moreover, the best method ('DNN_PCM') performed significantly better at almost one standard deviation higher than the mean performance. Furthermore, Multi-task and PCM implementations were shown to improve performance over single task Deep Neural Networks. Conversely, target prediction performed almost two standard deviations under the mean performance. Random Forests, Support Vector Machines, and Logistic Regression performed around mean performance. Finally, using an ensemble of DNNs, alongside additional tuning, enhanced the relative performance by another 27% (compared with unoptimized 'DNN_PCM'). Here, a standardized set to test and evaluate different machine learning algorithms in the context of multi-task learning is offered by providing the data and the protocols. Graphical Abstract .

Computational-experimental approach to drug-target interaction mapping: A case study on kinase inhibitors

Due to relatively high costs and labor required for experimental profiling of the full target space of chemical compounds, various machine learning models have been proposed as cost-effective means to advance this process in terms of predicting the most potent compound-target interactions for subsequent verification. However, most of the model predictions lack direct experimental validation in the laboratory, making their practical benefits for drug discovery or repurposing applications largely unknown. Here, we therefore introduce and carefully test a systematic computational-experimental framework for the prediction and pre-clinical verification of drug-target interactions using a well-established kernel-based regression algorithm as the prediction model. To evaluate its performance, we first predicted unmeasured binding affinities in a large-scale kinase inhibitor profiling study, and then experimentally tested 100 compound-kinase pairs. The relatively high correlation of 0.77 (p < 0.0001) between the predicted and measured bioactivities supports the potential of the model for filling the experimental gaps in existing compound-target interaction maps. Further, we subjected the model to a more challenging task of predicting target interactions for such a new candidate drug compound that lacks prior binding profile information. As a specific case study, we used tivozanib, an investigational VEGF receptor inhibitor with currently unknown off-target profile. Among 7 kinases with high predicted affinity, we experimentally validated 4 new off-targets of tivozanib, namely the Src-family kinases FRK and FYN A, the non-receptor tyrosine kinase ABL1, and the serine/threonine kinase SLK. Our sub-sequent experimental validation protocol effectively avoids any possible information leakage between the training and validation data, and therefore enables rigorous model validation for practical applications. These results demonstrate that the kernel-based modeling approach offers practical benefits for probing novel insights into the mode of action of investigational compounds, and for the identification of new target selectivities for drug repurposing applications.

Progress towards a public chemogenomic set for protein kinases and a call for contributions

Protein kinases are highly tractable targets for drug discovery. However, the biological function and therapeutic potential of the majority of the 500+ human protein kinases remains unknown. We have developed physical and virtual collections of small molecule inhibitors, which we call chemogenomic sets, that are designed to inhibit the catalytic function of almost half the human protein kinases. In this manuscript we share our progress towards generation of a comprehensive kinase chemogenomic set (KCGS), release kinome profiling data of a large inhibitor set (Published Kinase Inhibitor Set 2 (PKIS2)), and outline a process through which the community can openly collaborate to create a KCGS that probes the full complement of human protein kinases.

Profile-QSAR 2.0: Kinase Virtual Screening Accuracy Comparable to Four-Concentration IC50s for Realistically Novel Compounds

While conventional random forest regression (RFR) virtual screening models appear to have excellent accuracy on random held-out test sets, they prove lacking in actual practice. Analysis of 18 historical virtual screens showed that random test sets are far more similar to their training sets than are the compounds project teams actually order. A new, cluster-based "realistic" training/test set split, which mirrors the chemical novelty of real-life virtual screens, recapitulates the poor predictive power of RFR models in real projects. The original Profile-QSAR (pQSAR) method greatly broadened the domain of applicability over conventional models by using as independent variables a profile of activity predictions from all historical assays in a large protein family. However, the accuracy still fell short of experiment on realistic test sets. The improved "pQSAR 2.0" method replaces probabilities of activity from naïve Bayes categorical models at several thresholds with predicted IC₅₀s from RFR models. Unexpectedly, the high accuracy also requires removing the RFR model for the actual assay of interest from the independent variable profile. With these improvements, pQSAR 2.0 activity predictions are now statistically comparable to medium-throughput four-concentration IC₅₀ measurements even on the realistic test set. Beyond the yes/no activity predictions from a typical high-throughput screen (HTS) or conventional virtual screen, these semiquantitative IC₅₀ predictions allow for predicted potency, ligand efficiency, lipophilic efficiency, and selectivity against antitargets, greatly facilitating hitlist triaging and enabling virtual screening panels such as toxicity panels and overall promiscuity predictions.

Kinome-Wide Profiling Prediction of Small Molecules

Extensive kinase profiling data, covering more than half of the human kinome, are available nowadays and allow the construction of activity prediction models of high practical utility. Proteochemometric (PCM) approaches use compound and protein descriptors, which enables the extrapolation of bioactivity values to thus far unexplored kinases. In this study, the potential of PCM to make large-scale predictions on the entire kinome is explored, considering the applicability on novel compounds and kinases, including clinically relevant mutants. A rigorous validation indicates high predictive power on left-out kinases and superiority over individual kinase QSAR models for new compounds. Furthermore, external validation on clinically relevant mutant kinases reveals an excellent predictive power for mutations spread across the ATP binding site.