Search

Antibacterial drugs (AD) change the metabolic status of bacteria, contributing to bacterial death. However, antibiotic resistance and the emergence of multidrug-resistant bacteria increase interest in understanding metabolic network (MN) mutations and the interaction of AD vs MN. In this study, we employed the IFPTML = Information Fusion (IF) + Perturbation Theory (PT) + Machine Learning (ML) algorithm on a huge dataset from the ChEMBL database, which contains >155,000 AD assays vs >40 MNs of multiple bacteria species. We built a linear discriminant analysis (LDA) and 17 ML models centered on the linear index and based on atoms to predict antibacterial compounds. The IFPTML-LDA model presented the following results for the training subset: specificity (Sp) = 76% out of 70,000 cases, sensitivity (Sn) = 70%, and Accuracy (Acc) = 73%. The same model also presented the following results for the validation subsets: Sp = 76%, Sn = 70%, and Acc = 73.1%. Among the IFPTML nonlinear models, the k nearest neighbors (KNN) showed the best results with Sn = 99.2%, Sp = 95.5%, Acc = 97.4%, and Area Under Receiver Operating Characteristic (AUROC) = 0.998 in training sets. In the validation series, the Random Forest had the best results: Sn = 93.96% and Sp = 87.02% (AUROC = 0.945). The IFPTML linear and nonlinear models regarding the ADs vs MNs have good statistical parameters, and they could contribute toward finding new metabolic mutations in antibiotic resistance and reducing time/costs in antibacterial drug research., G.D.H. acknowledges financial support from grants from the Ministry of Science and Innovation (PID 2019-104148 GB-I00) and grant no. IT1045-16-2016–2021 from the Basque Government.

The aim of scaffold hopping (SH) is to find compounds consisting of different scaffolds from those in already known active compounds, giving an opportunity for unexplored regions of chemical space. We previously demonstrated the usefulness of pharmacophore graphs (PhGs) for this purpose through proof-of-concept virtual screening experiments. PhGs consist of nodes and edges corresponding to pharmacophoric features (PFs) and their topological distances. Although PhGs were effective in SH, they are hard to interpret as they are complete graphs. Herein, we introduce an intuitive representation of a molecule, termed as sparse pharmacophore graphs (SPhG) by keeping the topological distances among PFs as much as possible while reducing the number of edges in the graphs. Several benchmark calculations quantitatively confirmed the sparseness of the graphs and the preservation of topological distances among pharmacophoric points. As proof-of-concept applications, virtual screening (VS) trials for SH were conducted using active and inactive compounds from ChEMBL and PubChem databases for three biological targets: thrombin, tyrosine kinase ABL1, and κ-opioid receptor. The performances of VS were comparable with using fully connected PhGs. Furthermore, highly ranked SPhGs were interpretable for the three biological targets, in particular for thrombin, for which selected SPhGs were in agreement with the structure-based interpretation.

pKa is an important property in the lead optimization process since the charge state of a molecule in physiologic pH plays a critical role in its biological activity, solubility, membrane permeability, metabolism, and toxicity. Accurate and fast estimation of small molecule pKa is vital during the drug discovery process. We present MolGpKa, a web server for pKa prediction using a graph-convolutional neural network model. The model works by learning pKa related chemical patterns automatically and building reliable predictors with learned features. ACD/pKa data for 1.6 million compounds from the ChEMBL database was used for model training. We found that the performance of the model is better than machine learning models built with human-engineered fingerprints. Detailed analysis shows that the substitution effect on pKa is well learned by the model. MolGpKa is a handy tool for the rapid estimation of pKa during the ligand design process. The MolGpKa server is freely available to researchers and can be accessed at https://xundrug.cn/molgpka.

Physicochemical descriptors commonly used to define "drug-likeness" and ligand efficiency measures are assessed for their ability to differentiate marketed drugs from compounds reported to bind to their efficacious target or targets. Using ChEMBL version 26, a data set of 643 drugs acting on 271 targets was assembled, comprising 1104 drug-target pairs having ≥100 published compounds per target. Taking into account changes in their physicochemical properties over time, drugs are analyzed according to their target class, therapy area, and route of administration. Recent drugs, approved in 2010-2020, display no overall differences in molecular weight, lipophilicity, hydrogen bonding, or polar surface area from their target comparator compounds. Drugs are differentiated from target comparators by higher potency, ligand efficiency (LE), lipophilic ligand efficiency (LLE), and lower carboaromaticity. Overall, 96% of drugs have LE or LLE values, or both, greater than the median values of their target comparator compounds.

The SureChEMBL database provides open access to 17 million chemical entities mentioned in 14 million patents published since 1970. However, alongside with molecules covered by patent claims, the database is full of starting materials and intermediate products of little pharmacological relevance. Herein, we introduce a new filtering protocol to automatically select the core chemical structures best representing a congeneric series of pharmacologically relevant molecules in patents. The protocol is first validated against a selection of 890 SureChEMBL patents for which a total of 51,738 manually curated molecules are deposited in ChEMBL. Our protocol was able to select 92.5% of the molecules in ChEMBL from all 270,968 molecules in SureChEMBL for those patents. Subsequently, the protocol was applied to all 240,988 US pharmacological patents for which 9,111,706 molecules are available in SureChEMBL. The unsupervised filtering process selected 5,949,214 molecules (65.3% of the total number of molecules) that form highly congeneric chemical series in 188,795 of those patents (78.3% of the total number of patents). A SureChEMBL version enriched with molecules of pharmacological relevance is available for download at https://ftp.ebi.ac.uk/pub/databases/chembl/SureChEMBLccs. This work was supported by a RETOS project from the Spanish Ministerio de Ciencia, Innovación y Universidades (SAF2017‐83614‐R).

The safety of marketed drugs is an ongoing concern, with some of the more frequently prescribed medicines resulting in serious or life-threatening adverse effects in some patients. Safety-related information for approved drugs has been curated to include the assignment of toxicity class(es) based on their withdrawn status and/or black box warning information described on medicinal product labels. The ChEMBL resource contains a wide range of bioactivity data types, from early "Discovery" stage preclinical data for individual compounds through to postclinical data on marketed drugs; the inclusion of the curated drug safety data set within this framework can support a wide range of safety-related drug discovery questions. The curated drug safety data set will be made freely available through ChEMBL and updated in future database releases.

Machine learning methods are attracting considerable attention from the pharmaceutical industry for use in drug discovery and applications beyond. In recent studies, we and others have applied multiple machine learning algorithms and modeling metrics and, in some cases, compared molecular descriptors to build models for individual targets or properties on a relatively small scale. Several research groups have used large numbers of datasets from public databases such as ChEMBL in order to evaluate machine learning methods of interest to them. The largest of these types of studies used on the order of 1400 datasets. We have now extracted well over 5000 datasets from CHEMBL for use with the ECFP6 fingerprint and in comparison of our proprietary software Assay Central with random forest, k-nearest neighbors, support vector classification, naïve Bayesian, AdaBoosted decision trees, and deep neural networks (three layers). Model performance was assessed using an array of fivefold cross-validation metrics including area-under-the-curve, F1 score, Cohen's kappa, and Matthews correlation coefficient. Based on ranked normalized scores for the metrics or datasets, all methods appeared comparable, while the distance from the top indicated that Assay Central and support vector classification were comparable. Unlike prior studies which have placed considerable emphasis on deep neural networks (deep learning), no advantage was seen in this case. If anything, Assay Central may have been at a slight advantage as the activity cutoff for each of the over 5000 datasets representing over 570,000 unique compounds was based on Assay Central performance, although support vector classification seems to be a strong competitor. We also applied Assay Central to perform prospective predictions for the toxicity targets PXR and hERG to further validate these models. This work appears to be the largest scale comparison of these machine learning algorithms to date. Future studies will likely evaluate additional databases, descriptors, and machine learning algorithms and further refine the methods for evaluating and comparing such models.

The rise of novel artificial intelligence (AI) methods necessitates their benchmarking against classical machine learning for a typical drug-discovery project. Inhibition of the potassium ion channel, whose alpha subunit is encoded by the human ether-a-go-go-related gene (hERG), leads to a prolonged QT interval of the cardiac action potential and is a significant safety pharmacology target for the development of new medicines. Several computational approaches have been employed to develop prediction models for the assessment of hERG liabilities of small molecules including recent work using deep learning methods. Here, we perform a comprehensive comparison of hERG effect prediction models based on classical approaches (random forests and gradient boosting) and modern AI methods [deep neural networks (DNNs) and recurrent neural networks (RNNs)]. The training set (∼9000 compounds) was compiled by integrating the hERG bioactivity data from the ChEMBL database with experimental data generated from an in-house, high-throughput thallium flux assay. We utilized different molecular descriptors including the latent descriptors, which are real-value continuous vectors derived from chemical autoencoders trained on a large chemical space (>1.5 million compounds). The models were prospectively validated on ∼840 in-house compounds screened in the same thallium flux assay. The best results were obtained with the XGBoost method and RDKit descriptors. The comparison of models based only on latent descriptors revealed that the DNNs performed significantly better than the classical methods. The RNNs that operate on SMILES provided the highest model sensitivity. The best models were merged into a consensus model that offered superior performance compared to reference models from academic and commercial domains. Furthermore, we shed light on the potential of AI methods to exploit the big data in chemistry and generate novel chemical representations useful in predictive modeling and tailoring a new chemical space.

Mining the steadily increasing amount of chemical and biological data is a key challenge in drug discovery. Graph databases offer viable alternatives for capturing interrelationships between molecules and for generating novel insights for design. In a graph database, molecules and their properties are mapped to nodes, while relationships are described by edges. Here, we introduce a graph database for navigation in chemical space, analogue searching, and structure-activity relationship (SAR) analysis. We illustrate this concept using hERG channel inhibitors from ChEMBL to extract SAR knowledge. This graph database is built using different relationships, namely 2D-fingerprint similarity, matched molecular pairs, topomer distances, and structure-activity landscape indices (SALI). Typical applications include retrieving analogues linked by single or multiple edge paths to the query compound as well as detection of nonadditive SAR features. Finally, we identify triplets of linked molecules for clustering. The speed of searching and analysis allows the user to interactively navigate the database and to address complex questions in real-time.

Alzheimer’s disease (AD) is the most common cause of dementia, affecting approximately 35 million people worldwide. The current treatment options for people with AD consist of drugs designed to slow the rate of decline in memory and cognition, but these treatments are not curative, and patients eventually suffer complete cognitive injury. With the substantial amounts of published data on targets for this disease, we proposed that machine learning software could be used to find novel small-molecule treatments that can supplement the AD drugs currently on the market. In order to do this, we used publicly available data in ChEMBL to build and validate Bayesian machine learning models for AD target proteins. The first AD target that we have addressed with this method is the serine–threonine kinase glycogen synthase kinase 3 beta (GSK3β), which is a proline-directed serine–threonine kinase that phosphorylates the microtubule-stabilizing protein tau. This phosphorylation prompts tau to dissociate from the microtubule and form insoluble oligomers called paired helical filaments, which are one of the components of the neurofibrillary tangles found in AD brains. Using our Bayesian machine learning model for GSK3β consisting of 2368 molecules, this model produced a five-fold cross validation ROC of 0.905. This model was also used for virtual screening of large libraries of FDA-approved drugs and clinical candidates. Subsequent testing of selected compounds revealed a selective small-molecule inhibitor, ruboxistaurin, with activity against GSK3β (avg IC50 = 97.3 nM) and GSK3α (IC50 = 695.9 nM). Several other structurally diverse inhibitors were also identified. We are now applying this machine learning approach to additional AD targets to identify approved drugs or clinical trial candidates that can be repurposed as AD therapeutics. This represents a viable approach to accelerate drug discovery and do so at a fraction of the cost of traditional high throughput screening.

Protein-protein interactions (PPIs) are attractive targets for drug design because of their essential role in numerous cellular processes and disease pathways. However, in general, PPIs display exposed binding pockets at the interface, and as such, have been largely unexploited for therapeutic interventions with low-molecular weight compounds. Here, we used docking and various rescoring strategies in an attempt to recover PPI inhibitors from a set of active and inactive molecules for 11 targets collected in ChEMBL and PubChem. Our focus is on the screening power of the various developed protocols and on using fast approaches so as to be able to apply such a strategy to the screening of ultralarge libraries in the future. First, we docked compounds into each target using the fast "pscreen" mode of the structure-based virtual screening (VS) package Surflex. Subsequently, the docking poses were postprocessed to derive a set of 3D topological descriptors: (i) shape similarity and (ii) interaction fingerprint similarity with a co-crystallized inhibitor, (iii) solvent-accessible surface area, and (iv) extent of deviation from the geometric center of a reference inhibitor. The derivatized descriptors, together with descriptor-scaled scoring functions, were utilized to investigate possible impacts on VS performance metrics. Moreover, four standalone scoring functions, RF-Score-VS (machine-learning), DLIGAND2 (knowledge-based), Vinardo (empirical), and X-SCORE (empirical), were employed to rescore the PPI compounds. Collectively, the results indicate that the topological scoring algorithms could be valuable both at a global level, with up to 79% increase in areas under the receiver operating characteristic curve for some targets, and in early stages, with up to a 4-fold increase in enrichment factors at 1% of the screened collections. Outstandingly, DLIGAND2 emerged as the best scoring function on this data set, outperforming all rescoring techniques in terms of VS metrics. The described methodology could help in the rational design of small-molecule PPI inhibitors and has direct applications in many therapeutic areas, including cancer, CNS, and infectious diseases such as COVID-19.

Nanosystems are gaining momentum in pharmaceutical sciences because of the wide variety of possibilities for designing these systems to have specific functions. Specifically, studies of new cancer cotherapy drug-vitamin release nanosystems (DVRNs) including anticancer compounds and vitamins or vitamin derivatives have revealed encouraging results. However, the number of possible combinations of design and synthesis conditions is remarkably high. In addition, a large number of anticancer and vitamin derivatives have been already assayed, but a notably less number of cases of DVRNs were assayed as a whole (with the anticancer compound and the vitamin linked to them). Our approach combines with the perturbation theory and machine learning (PTML) model to predict the probability of obtaining an interesting DVRN by changing the anticancer compound and/or the vitamin present in a DVRN that is already tested for other anticancer compounds or vitamins that have not been tested yet as part of a DVRN. In a previous work, we built a linear PTML model useful for the design of these nanosystems. In doing so, we used information fusion (IF) techniques to carry out data enrichment of DVRN data compiled from the literature with the data for preclinical assays of vitamins from the ChEMBL database. The design features of DVRNs and the assay conditions of nanoparticles (NPs) and vitamins were included as multiplicative PT operators (PTOs) to the system, which indicates the importance of these variables. However, the previous work omitted experiments with nonlinear ML techniques and different types of PTOs such as metric-based PTOs. More importantly, the previous work does not consider the structure of the anticancer drug to be included in the new DVRNs. In this work, we are going to accomplish three main objectives (tasks). In the first task, we found a new model, alternative to the one published before, for the rational design of DVRNs using metric-based PTOs. The most accurate PTML model was the artificial neural network model, which showed values of specificity, sensitivity, and accuracy in the range of 90-95% in training and external validation series for more than 130,000 cases (DVRNs vs ChEMBL assays). Furthermore, in the second task, we used IF techniques to carry out data enrichment of our previous data set. In doing so, we constructed a new working data set of >970,000 cases with the data of preclinical assays of DVRNs, vitamins, and anticancer compounds from the ChEMBL database. All these assays have multiple continuous variables or descriptors dk and categorical variables cj (conditions of the assays) for drugs (dack, cacj), vitamins (dvk, cvj), and NPs (dnk, cnj). These data include >20,000 potential anticancer compounds with >270 protein targets (cac1), >580 assay cell organisms (cac2), and so forth. Furthermore, we include >36,000 assay vitamin derivatives in >6200 types of cells (c2vit), >120 assay organisms (c3vit), >60 assay strains (c4vit), and so forth. The enriched data set also contains >20 types of DVRNs (c5n) with 9 NP core materials (c4n), 8 synthesis methods (c7n), and so forth. We expressed all this information with PTOs and developed a qualitatively new PTML model that incorporates information of the anticancer drugs. This new model presents 96-97% of accuracy for training and external validation subsets. In the last task, we carried out a comparative study of ML and/or PTML models published and described how the models we are presenting cover the gap of knowledge in terms of drug delivery. In conclusion, we present here for the first time a multipurpose PTML model that is able to select NPs, anticancer compounds, and vitamins and their conditions of assay for DVRN design.

Although the NaV1.7 sodium channel is a promising drug target for pain, traditional screening strategies for discovery of NaV1.7 inhibitors are very painstaking and time-consuming. Herein, we aimed to build machine learning models for screening and design of potent and effective NaV1.7 sodium channel inhibitors. We customized the imbalanced data set from ChEMBL and BindingDB to train and filter the best classification model. Then, the whole-cell voltage-clamp was employed to validate the inhibitors. We assembled a molecular group optimization method by combining the Grammar Variational Autoencoder, classification model, and simulated annealing. We found that the RF-CDK model (random forest + CDK fingerprint) performs best in the imbalanced data set. Of the three compounds that may have inhibitory effects, nortriptyline has been experimentally verified. In the molecule optimization process, 40 molecules located in the applicability domain of RF-CDK were used as a starting point, among which 34 molecules evolved to molecules with greater molecular scores (MS). The molecule with the highest MS was derived from CHEMBL2325245. The model and method we developed for NaV1.7 inhibitors are also applicable to other targets.

Despite the well-known adverse health effects associated with tobacco use, addiction to nicotine found in tobacco products causes difficulty in quitting among users. Nicotinic acetylcholine receptors (nAChRs) are the physiological targets of nicotine and facilitate addiction to tobacco products. The nAChR-α7 subtype plays an important role in addiction; therefore, predicting the binding activity of tobacco constituents to nAChR-α7 is an important component for assessing addictive potential of tobacco constituents. We developed an α7 binding activity prediction model based on a large training data set of 843 chemicals with human α7 binding activity data extracted from PubChem and ChEMBL. The model was tested using 1215 chemicals with rat α7 binding activity data from the same databases. Based on the competitive docking results, the docking scores were partitioned to the key residues that play important roles in the receptor-ligand binding. A decision forest was used to train the human α7 binding activity prediction model based on the partition of docking scores. Five-fold cross validations were conducted to estimate the performance of the decision forest models. The developed model was used to predict the potential human α7 binding activity for 5275 tobacco constituents. The human α7 binding activity data for 84 of the 5275 tobacco constituents were experimentally measured to confirm and empirically validate the prediction results. The prediction accuracy, sensitivity, and specificity were 64.3, 40.0, and 81.6%, respectively. The developed prediction model of human α7 may be a useful tool for high-throughput screening of potential addictive tobacco constituents.

Determining the biological activity of vitamin derivatives is needed given that organic synthesis of analogs of vitamins is an active field of interest for medicinal chemistry, pharmaceuticals, and food additives. Accordingly, scientists from different disciplines perform preclinical assays (nij) with a considerable combination of assay conditions (cj). Indeed, the ChEMBL platform contains a database that includes results from 36 220 different biological activity bioassays of 21 240 different vitamins and vitamin derivatives. These assays present are heterogeneous in terms of assay combinations of cj. They are focused on >500 different biological activity parameters (c0), >340 different targets (c1), >6200 types of cell (c2), >120 organisms of assay (c3), and >60 assay strains (c4). It includes a total of >1850 niacin assays, >1580 tretinoin assays, >1580 retinol assays, 857 ascorbic acid assays, etc. Given the complexity of this combinatorial data in terms of being assimilated by researchers, we propose to build a model by combining perturbation theory (PT) and machine learning (ML). Through this study, we propose a PTML (PT + ML) combinatorial model for ChEMBL results on biological activity of vitamins and vitamins derivatives. The linear discriminant analysis (LDA) model presented the following results for training subset a: specificity (%) = 90.38, sensitivity (%) = 87.51, and accuracy (%) = 89.89. The model showed the following results for the external validation subset: specificity (%) = 90.58, sensitivity (%) = 87.72, and accuracy (%) = 90.09. Different types of linear and nonlinear PTML models, such as logistic regression (LR), classification tree (CT), naive Bayes (NB), and random Forest (RF), were applied to contrast the capacity of prediction. The PTML-LDA model predicts with more accuracy by applying combinatorial descriptors. In addition, a PCA experiment with chemical structure descriptors allowed us to characterize the high structural diversity of the chemical space studied. In any case, PTML models using chemical structure descriptors do not improve the performance of the PTML-LDA model based on ALOGP and PSA. We can conclude that the three variable PTML-LDA model is a simplified and adaptable tool for the prediction, for different experiment combinations, the biological activity of derivative vitamins.

A new methodology for classifying fragment combinations and characterizing pseudo-natural products (PNPs) is described. The source code is based on open-source tools and is organized as a Python package. Tasks can be executed individually or within the context of scalable, robust workflows. First, structures are standardized and duplicate entries are filtered out. Then, molecules are probed for the presence of predefined fragments. For molecules with more than one match, fragment combinations are classified. The algorithm considers the pair-wise relative position of fragments within the molecule (fused atoms, linkers, intermediary rings), resulting in 18 different possible fragment combination categories. Finally, all combinations for a given molecule are assembled into a fragment combination graph, with fragments as nodes and combination types as edges. This workflow was applied to characterize PNPs in the ChEMBL database via comparison of fragment combination graphs with Natural Product (NP) references, represented by the Dictionary of Natural Products. The Murcko fragments extracted from 2,000 structures previously described were used to define NP-fragments. The results indicate that ca. 23% of the biologically relevant compounds listed in ChEMBL comply to the PNP definition, and that, therefore, PNPs occur frequently among known biologically relevant small molecules. The majority (>95%) of PNPs contains two to four fragments, mainly (>95%) distributed in five different combination types. These findings may provide guidance for the design of new PNPs.

Generative neural networks have shown promising results in de novo drug design. Recent studies suggest that one of the efficient ways to produce novel molecules matching target properties is to model SMILES sequences using deep learning in a way similar to language modeling in natural language processing. In this paper, we present a survey of various machine learning methods for SMILES-based language modeling and propose our benchmarking results on a standardized subset of ChEMBL database.

Discovery of new antibacterial agents is a never-ending task of medicinal chemistry. Every new drug brings significant improvement to patients with bacterial infections, but prolonged usage of antibacterials leads to the emergence of resistant strains. Therefore, novel active structures with new modes of action are required. We describe a web application called AntiBac-Pred aimed to help users in the rational selection of the chemical compounds for experimental studies of antibacterial activity. This application is developed using antibacterial activity data available in ChEMBL and PASS software. It allows users to classify chemical structures of interest into growth inhibitors or noninhibitors of 353 different bacteria strains, including both resistant and nonresistant ones.

G-protein-coupled receptors (GPCRs), also known as 7-transmembrane receptors, are the single largest class of drug targets. Consequently, a large amount of preclinical assays having GPCRs as molecu...

Retroviral infections, such as HIV, are, until now, diseases with no cure. Medicine and pharmaceutical chemistry need and consider it a huge goal to define target proteins of new antiretroviral compounds. ChEMBL manages Big Data features with a complex data set, which is hard to organize. This makes information difficult to analyze due to a big number of characteristics described in order to predict new drug candidates for retroviral infections. For this reason, we propose to develop a new predictive model combining perturbation theory (PT) bases and machine learning (ML) modeling to create a new tool that can take advantage of all the available information. The PTML model proposed in this work for the ChEMBL data set preclinical experimental assays for antiretroviral compounds consists of a linear equation with four variables. The PT operators used are founded on multicondition moving averages, combining different features and simplifying the difficulty to manage all data. More than 140 000 preclinical assays for 56 105 compounds with different characteristics or experimental conditions have been carried out and can be found in ChEMBL database, covering combinations with 359 biological activity parameters (c0), 55 protein accessions (c1), 83 cell lines (c2), 64 organisms of assay (c3), and 773 subtypes or strains. We have included 150 148 preclinical experimental assays for HIV virus, 1188 for HTLV virus, 84 for simian immunodeficiency virus, 370 for murine leukemia virus, 119 for Rous sarcoma virus, 1581 for MMTV, etc. We also included 5277 assays for hepatitis B virus. The developed PTML model reached considerable values in sensibility (73.05% for training and 73.10% for validation), specificity (86.61% for training and 87.17% for validation), and accuracy (75.84% for training and 75.98% for validation). We also compared alternative PTML models with different PT operators such as covariance, moments, and exponential terms. Finally, we made a comparison between literature ML models with our PTML model and also artificial neural network (ANN) nonlinear models. We conclude that this PTML model is the first one to consider multiple characteristics of preclinical experimental antiretroviral assays combined, generating a simple, useful, and adaptable instrument, which could reduce time and costs in antiretroviral drugs research.

ChEMBL biological activities prediction for 1-5-bromofur-2-il-2-bromo-2-nitroethene (G1) is a difficult task for cytokine immunotoxicity. The current study presents experimental results for G1 interaction with mouse Th1/Th2 and pro-inflammatory cytokines using a cytometry bead array (CBA). In the in vitro test of CBA, the results show no significant differences between the mean values of the Th1/Th2 cytokines for the samples treated with G1 with respect to the negative control, but there are moderate differences for cytokine values between different periods (24/48 h). The experiments show no significant differences between the mean values of the pro-inflammatory cytokines for the samples treated with G1, regarding the negative control, except for the values of tumor necrosis factor (TNF) and Interleukin (IL6) between the group treated with G1 and the negative control at 48 h. Differences occur for these cytokines in the periods (24/48 h). The study confirmed that the antimicrobial G1 did not alter the Th1/Th2 cytokines concentration in vitro in different periods, but it can alter TNF and IL6. G1 promotes free radicals production and activates damage processes in macrophages culture. In order to predict all ChEMBL activities for drugs in other experimental conditions, a ChEMBL data set was constructed using 25 biological activities, 1366 assays, 2 assay types, 4 assay organisms, 2 organisms, and 12 cytokine targets. Molecular descriptors calculated with Rcpi and 15 machine learning methods were used to find the best model able to predict if a drug could be active or not against a specific cytokine, in specific experimental conditions. The best model is based on 120 selected molecular descriptors and a deep neural network with area under the curve of the receiver operating characteristic of 0.904 and accuracy of 0.832. This model predicted 1384 G1 biological activities against cytokines in all ChEMBL data set experimental conditions.

Several recent reports have shown that long short-term memory generative neural networks (LSTM) of the type used for grammar learning efficiently learn to write Simplified Molecular Input Line Entry System (SMILES) of druglike compounds when trained with SMILES from a database of bioactive compounds such as ChEMBL and can later produce focused sets upon transfer learning with compounds of specific bioactivity profiles. Here we trained an LSTM using molecules taken either from ChEMBL, DrugBank, commercially available fragments, or from FDB-17 (a database of fragments up to 17 atoms) and performed transfer learning to a single known drug to obtain new analogs of this drug. We found that this approach readily generates hundreds of relevant and diverse new drug analogs and works best with training sets of around 40,000 compounds as simple as commercial fragments. These data suggest that fragment-based LSTM offer a promising method for new molecule generation.

Numerous studies have been published in recent years with acceptable quantitative structure-activity relationship (QSAR) modeling based on heterogeneous data. In many cases, the training sets for QSAR modeling were constructed from compounds tested by different biological assays, contradicting the opinion that QSAR modeling should be based on the data measured by a single protocol. We attempted to develop approaches that help to determine how heterogeneous data should be used for the creation of QSAR models on the basis of different sets of compounds tested by different experimental methods for the same target and the same endpoint. To this end, more than 100 QSAR models for the IC50 values of ligands interacting with cyclooxygenase 1,2 (COX) and seed lipoxygenase (LOX), obtained from ChEMBL database were created using the GUSAR software. The QSAR models were tested on the external set, including 26 new thiazolidinone derivatives, which were experimentally tested for COX-1,2/LOX inhibition. The IC50 values of the derivatives varied from 89 μM to 26 μM for LOX, from 200 μM to 0.018 μM for COX-1, and from 210 μM to 1 μM for COX-2. This study showed that the accuracy of the models is dependent on the distribution of IC50 values of low activity compounds in the training sets. In the most cases, QSAR models created based on the combined training sets had advantages in comparison with QSAR models, based on a single publication. We introduced a new method of combination of quantitative data from different experimental studies based on the data of reference compounds, which was called "scaling".

Here we report PPB2 as a target prediction tool assigning targets to a query molecule based on ChEMBL data. PPB2 computes ligand similarities using molecular fingerprints encoding composition (MQN), molecular shape and pharmacophores (Xfp), and substructures (ECfp4) and features an unprecedented combination of nearest neighbor (NN) searches and Naı̈ve Bayes (NB) machine learning, together with simple NN searches, NB and Deep Neural Network (DNN) machine learning models as further options. Although NN(ECfp4) gives the best results in terms of recall in a 10-fold cross-validation study, combining NN searches with NB machine learning provides superior precision statistics, as well as better results in a case study predicting off-targets of a recently reported TRPV6 calcium channel inhibitor, illustrating the value of this combined approach. PPB2 is available to assess possible off-targets of small molecule drug-like compounds by public access at http://gdb.unibe.ch .

We present MuSSeL, a multifingerprint similarity search algorithm, able to predict putative drug targets for a given query small molecule as well as to return a quantitative assessment of its bioactivity in terms of Ki or IC50 values. Predictions are automatically made exploiting a large collection of high quality experimental bioactivity data available from ChEMBL (version 22.1) combining, in a consensus-like approach, predictions resulting from a similarity search performed using 13 different fingerprint definitions. Importantly, the herein proposed algorithm is also effective in detecting and handling activity cliffs. A calibration set including small molecules present in the last updated version of ChEMBL (version 23) was employed to properly tune the algorithm parameters. Three randomly built external sets were instead challenged for model performances. The potential use of MuSSeL was also challenged by a prospective exercise for the prediction of five bioactive compounds taken from articles published in the Journal of Medicinal Chemistry just few months ago. The paper emphasizes the importance of implementing multifingerprint consensus strategies to increase the confidence in prediction of similarity search algorithms and provides a fast and easy-to-run tool for drug target and bioactivity prediction. Copyright: CC BY-NC 4.0

HIV-1 protease plays an important role in the processing of virus infection. Protease is an effective therapeutic target for the treatment of HIV-1. Our data set is based on a selection of 4855 HIV-1 protease inhibitors (PIs) from ChEMBL. A series of 15 classification models for predicting the active inhibitors were built by machine learning methods, including k-nearest neighors (K-NN), decision tree (DT), random forest (RF), support vector machine (SVM), and deep neural network (DNN). The molecular structures were characterized by (1) fingerprint descriptors including MACCS fingerprints and PubChem fingerprints and (2) physicochemical descriptors calculated by CORINA Symphony. The prediction accuracies of all of the models are more than 70% on the test set; the best accuracy of 83.07% was obtained by model 4A, which was built by the SVM method based on MACCS fingerprint descriptors. Nine consensus models were built with three kinds of different descriptors, which combined all of the machine learning methods using the “consensus prediction”. Model C3a developed with MACCS fingerprint descriptors showed the highest accuracy on both training set (91.96%) and test set (83.15%). An external validation set including 35 989 compounds from DUD database and 239 active inhibitors from the recent literature was used to verify the performance of our model. The best prediction accuracy of 98.37% was obtained by model 3C, which was built by RF based on CORINA Symphony descriptors. In addition, from the analysis of molecular descriptors, it shows that the aromatic system and atoms related to hydrogen bonding provide important contributions to the bioactivity of PIs.

Natural products (NPs), being evolutionary selected over millions of years to bind to biological macromolecules, remain an important source of inspiration for medicinal chemists even after the advent of efficient drug discovery technologies such as combinatorial chemistry and high-throughput screening. Thus, there is a strong demand for efficient and user-friendly computational tools that allow to analyze large libraries of NPs. In this context, we present NP Navigator – a free, intuitive online tool for visualization and navigation through the chemical space of NPs and NP-like molecules[1] (https://infochm.chimie.unistra.fr/npnav/chematlas_userspace ). It is based on the hierarchical ensemble of more than 200 Generative Topographic Maps(GTM)[2], featuring NPs from the COlleCtion of Open NatUral producTs (COCONUT), bioactive compounds from ChEMBL and commercially available molecules from ZINC. NP Navigator allows to efficiently analyze different aspects of NPs - chemotype distribution, physicochemical properties, reported and/or predicted biological activity and commercial availability of NPs. Users are welcome not only to browse through hundreds of thousands of compounds from ZINC, ChEMBL and COCONUT but also project a several external molecules that play the role of “chemical trackers” allowing to trace particular chemotypes in the NP chemical space and detect analogs of the compound of interest. [1] Y. Zabolotna, P. Ertl, D. Horvath, F. Bonachera, G. Marcou, A. Varnek, NP Navigator: a New Look at the Natural Product Chemical Space, 2021.[2] C. M. Bishop, M. Svensen, C. K. I. Williams, Neural Comput. 1998, 10, 215-234.

Machine learning scoring functions for protein-ligand binding affinity have been found to consistently outperform classical scoring functions when trained and tested on crystal structures of bound protein-ligand complexes. However, it is less clear how these methods perform when applied to docked poses of complexes. We explore how the use of docked rather than crystallographic poses for both training and testing affects the performance of machine learning scoring functions. Using the PDBbind Core Sets as benchmarks, we show that the performance of a structure-based machine learning scoring function trained and tested on docked poses is lower than that of the same scoring function trained and tested on crystallographic poses. We construct a hybrid scoring function by combining both structure-based and ligand-based features, and show that its ability to predict binding affinity using docked poses is comparable to that of purely structure-based scoring functions trained and tested on crystal poses. We also present a new, freely available validation set-the Updated DUD-E Diverse Subset-for binding affinity prediction using data from DUD-E and ChEMBL. Despite strong performance on docked poses of the PDBbind Core Sets, we find that our hybrid scoring function sometimes generalizes poorly to a protein target not represented in the training set, demonstrating the need for improved scoring functions and additional validation benchmarks.

In Mycobacterium tuberculosis (MTB), the cell wall synthesis flavoenzyme decaprenylphosphoryl-β- d -ribose 2′-epimerase (DprE1) plays a crucial role in host pathogenesis, virulence, lethality and survival under stress. The emergence of different variants of drug resistant MTB are a major threat worldwide which essentially requires more effective new drug molecules with no major side effects. Here, we used structure based virtual screening of bioactive molecules from the ChEMBL database targeting DprE1, having bioactive 78,713 molecules known for anti-tuberculosis activity. An extensive molecular docking, binding affinity and pharmacokinetics profile filtering results in the selection four compounds, C5 (ChEMBL2441313), C6 (ChEMBL2338605), C8 (ChEMBL441373) and C10 (ChEMBL1607606) which may explore as potential drug candidates. The obtained results were validated with thirteen known DprE1 inhibitors. We further estimated the free-binding energy, solvation and entropy terms underlying the binding properties of DprE1-ligand interactions with the implication of MD simulation, MM/GBSA, MM/PBSA and MM/3D-RISM. Interestingly, we find that C6 shows the highest ΔG scores (−41.28 ± 3.51, −22.36 ± 3.17, −10.33 ± 5.70 kcal mol−1) in MM/GBSA, MM/PBSA and MM/3D-RISM assay, respectively. Whereas, the lowest ΔG scores (−35.31 ± 3.44, −13.67 ± 2.65, −3.40 ± 4.06 kcal mol−1) observed for CT319, the inhibitor co-crystallized with DprE1. Collectively, the results demonstrated that hit-molecules: C5, C6, C8 and C10 having better binding free energy and molecular affinity as compared to CT319. Thus, we proposed that selected compounds may be explored as lead molecules in MTB therapy.

Machine learning methods are attracting considerable attention from the pharmaceutical industry for use in drug discovery and applications beyond. In recent studies we have applied multiple machine learning algorithms, modeling metrics and in some cases compared molecular descriptors to build models for individual targets or properties on a relatively small scale. Several research groups have used large numbers of datasets from public databases such as ChEMBL in order to evaluate machine learning methods of interest to them. The largest of these types of studies used on the order of 1400 datasets. We have now extracted well over 5000 datasets from CHEMBL for use with the ECFP6 fingerprint and comparison of our proprietary software Assay CentralTM with random forest, k-Nearest Neighbors, support vector classification, naïve Bayesian, AdaBoosted decision trees, and deep neural networks (3 levels). Model performance was assessed using an array of five-fold cross-validation metrics including area-under-the-curve, F1 score, Cohen’s kappa and Matthews correlation coefficient. Based on ranked normalized scores for the metrics or datasets all methods appeared comparable while the distance from the top indicated Assay CentralTM and support vector classification were comparable. Unlike prior studies which have placed considerable emphasis on deep neural networks (deep learning), no advantage was seen in this case where minimal tuning was performed of any of the methods. If anything, Assay CentralTM may have been at a slight advantage as the activity cutoff for each of the over 5000 datasets representing over 570,000 unique compounds was based on Assay CentralTMperformance, but support vector classification seems to be a strong competitor. We also apply Assay CentralTM to prospective predictions for PXR and hERG to further validate these models. This work currently appears to be the largest comparison of machine learning algorithms to date. Future studies will likely evaluate additional databases, descriptors and algorithms, as well as further refining methods for evaluating and comparing models.

Protein kinases play a crucial role in many cell signaling processes, making them one of the most important families of drug targets. In this context, fragment-based drug design strategies have been successfully applied to develop novel kinase inhibitors, usually following a knowledge-driven approach to optimize a focused set of fragments to a potent kinase inhibitor. Alternatively, KinFragLib is a new method that allows to explore and extend the chemical space of kinase inhibitors using data-driven fragmentation and recombination, built on available structural kinome data from the KLIFS database for over 2,500 kinase DFG-in complexes. The computational fragmentation method splits the co-crystallized non-covalent kinase inhibitors into fragments with respect to their 3D proximity to six predefined functionally relevant subpocket centers. The resulting fragment library consists of six subpocket pools with over 7,000 fragments, available at https://github.com/volkamerlab/KinFragLib.KinFragLib offers two main applications: (i) In-depth analyses of the chemical space of known kinase inhibitors, subpocket characteristics and connections, as well as (ii) subpocket-informed recombination of fragments to generate potential novel inhibitors. The latter showed that recombining only a subset of 624 representative fragments generated a combinatorial library of 6.7 million molecules, containing, besides some known kinase inhibitors, more than 99% novel chemical matter compared to ChEMBL and 63% molecules compliant with Lipinski's rule of five.Note: This document is the unedited Author’s version of a Submitted Work that was subsequently accepted for publication in the Journal of Chemical Information and Modeling, copyright © American Chemical Society after peer review. To access the final edited and published work see https://pubs.acs.org/doi/abs/10.1021/acs.jcim.0c00839.

In silico drug design can play a vital role in identifying promising drug candidates against COVID-19. Herein, we focused on the main protease of SARS-CoV-2 that plays crucial biological functions in the virus. We performed a ligand-based virtual screening followed by a docking screening for testing approved drugs and bioactive compounds listed in the DrugBank and ChEMBL databases. The top 8 docking results were advanced to all-atom MD simulations to study the relative stability of the protein-ligand interactions.Our results suggest several promising approved and bioactive inhibitors of SARS-CoV-2 Mpro.

Simplified molecular input line entry system (SMILES)-based deep learning models are slowly emerging as an important research topic in cheminformatics. In this study, we introduce SMILES pair encoding (SPE), a data-driven tokenization algorithm. SPE first learns a vocabulary of high-frequency SMILES substrings from a large chemical dataset (e.g., ChEMBL) and then tokenizes SMILES based on the learned vocabulary for the actual training of deep learning models. SPE augments the widely used atom-level tokenization by adding human-readable and chemically explainable SMILES substrings as tokens. Case studies show that SPE can achieve superior performances on both molecular generation and quantitative structure-activity relationship (QSAR) prediction tasks. In particular, the SPE-based generative models outperformed the atom-level tokenization model in the aspects of novelty, diversity, and ability to resemble the training set distribution. The performance of SPE-based QSAR prediction models were evaluated using 24 benchmark datasets where SPE consistently either did match or outperform atom-level and k-mer tokenization. Therefore, SPE could be a promising tokenization method for SMILES-based deep learning models. An open-source Python package SmilesPE was developed to implement this algorithm and is now freely available at https://github.com/XinhaoLi74/SmilesPE.

Discovery of drugs against newly emerged pathogenic agents like the SARS-CoV-2 coronavirus (CoV) must be based on previous research against related species. Scientists need to get acquainted with and develop a global oversight over so-far tested molecules. Chemography (herein used Generative Topographic Mapping, in particular) places structures on a human-readable 2D map (obtained by dimensionality reduction of the chemical space of molecular descriptors) and is thus well suited for such an audit. The goal is to map medicinal chemistry efforts so far targeted against CoVs. This includes comparing libraries tested against various virus species/genera, predicting their polypharmacological profiles and highlighting often encountered chemotypes. Maps are challenged to provide predictive activity landscapes against viral proteins. Definition of “anti-CoV” map zones led to selection of therein residing 380 potential anti-CoV agents, out of a vast pool of 800M organic compounds.

Background: Molecular fingerprints are essential cheminformatics tools for virtual screening and mapping chemical space. Among the different types of fingerprints, substructure fingerprints perform best for small molecules such as drugs, while atom-pair fingerprints are preferable for large molecules such as peptides. However, no available fingerprint achieves good performance on both classes of molecules. Results: Here we set out to design a new fingerprint suitable for both small and large molecules by combining substructure and atom-pair concepts. Our quest resulted in a new fingerprint called MinHashed atom-pair fingerprint up to a diameter of four bonds (MAP4). In this fingerprint the circular substructures with radii of r = 1 and r = 2 bonds around each atom in an atom-pair are written as two pairs of SMILES, each pair being combined with the topological distance separating the two central atoms. These so-called atom-pair molecular shingles are hashed, and the resulting set of hashes is MinHashed to form the MAP4 fingerprint. MAP4 significantly outperforms all other fingerprints on an extended benchmark that combines the Riniker and Landrum small molecule benchmark with a peptide benchmark recovering BLAST analogs from either scrambled or point mutation analogs. MAP4 furthermore produces well-organized chemical space tree-maps (TMAPs) for databases as diverse as DrugBank, ChEMBL, SwissProt and the Human Metabolome Database (HMBD), and differentiates between all metabolites in HMBD, over 70 % of which are indistinguishable from their nearest neighbor using substructure fingerprints. Conclusion: MAP4 is a new molecular fingerprint suitable for drugs, biomolecules, and the metabolome and can be adopted as a universal fingerprint to describe and search chemical space. The source code is available at https://github.com/reymond-group/map4 and interactive MAP4 similarity search tools and TMAPs for various databases are accessible at http://map-search.gdb.tools/ and http://tm.gdb.tools/map4/.

This work introduces a novel chemical space embedding method "Hilbert-Curve Assisted Structure Embedding (HCASE)" with help of pseudo-Hilbert Curves and Scaffold- Keys. The method was designed to produce an embedding that can be intuitively interpreted by medicinal chemists and data analysts. We analyzed the embedding of approved drug molecules (DrugBank) and natural products (CANVASS) into chemical spaces defined by Bemis-Murcko scaffolds extracted from ChEMBL (v24.1) database and from ChEMBL (v23) Natural Products. The implementation of HCASE algorithm and the input and results files of the analyses are available at https://github.com/ncats/hcase .

We have developed a neural network potential energy function for use in drug discovery, with chemical element support extended from 41% to 94% of druglike molecules based on ChEMBL. We expand on the work of Smith et al., with their highly accurate network for the elements H, C, N, O, creating a network for H, C, N, O, S, F, Cl, P. We focus particularly on the calculation of relative conformer energies, for which we show that our new potential energy function has an RMSE of 0.70 kcal/mol for prospective druglike molecule conformers, substantially better than the previous state of the art. The speed and accuracy of this model could greatly accelerate the parameterization of protein-ligand binding free energy calculations for novel druglike molecules., 20 pages, 6 figures

Small molecule tool compounds have enabled profound advances in life science research. These chemicals are potent, cell active, and selective, and, thus, are suitable for interrogating biological processes. For these chemicals to be useful they must be correctly characterized and researchers must be aware of them. We mined the ChEMBL bioactivity database to identify high quality tool compounds in an unbiased way. We identified 407 best-in-class compounds for 278 protein targets, and these are reported in an annotated data set. Additionally, we developed informatics functions and a web application for data visualization and automated pharmacological hypothesis generation. These functions were used to predict inhibitors of the Chromobox Protein Homologue 5 (CBX5) mediated gene repression pathway that currently lacks appropriate inhibitors. The predictions were subsequently validated by a highly specific cell based assay, revealing new chemical modulators of CBX5-mediated heterochromatin formation. This data set and associated functions will help researchers make the best use of these valuable compounds.

Phosphodiesterases are proving to be fruitful targets for drug discovery. At the same time fragment-based drug discovery has matured into a powerful and widely applied technique. In this communication we review the application of fragment-based drug discovery for the successful identification of novel 3′,5′-cyclic nucleotide phosphodiesterase (PDE) inhibitors, concentrating on both experimental and computational strategies for fragment screening and hit-to-lead development. To this end, we also mine the open access databases ChEMBL and PDB for fragments showing PDE inhibitory activity, as well as SureChEMBL for recent PDE related patents, to provide a wider context for exploring fragment diversity. Together these approaches form an integrated experimental and computational platform to exploit fragment-based drug discovery for this important gene superfamily.

Design strategies centered on intramolecular hydrogen bonds are sometime used in drug discovery, but their general applicability has not been addressed beyond scattered examples or circumstantial evidence. A total of 1053 matched molecular pairs where only one of the two molecules is able to form an intramolecular hydrogen bond via monatomic transformations have been identified across the ChEMBL database. These pairs were used to investigate the effect of intramolecular hydrogen bonds on biological activity. While cases of extreme, conflicting variation of effect emerge, the mean biological activity difference for a pair is close to zero and does not exceed ±0.5 log biological activity for over 50% of the analyzed sample.

A more in-depth analysis of RISC shows that its reported performance for 1024- and 2048-bit fingerprints does not extend to many other fingerprint types of the same size. While faster than chemfp’s BitBound for the atypically sparse Morgan circular fingerprints, its performance is comparable to BitBound for other types with similar density, and significantly slower as bit densities increase above around 0.1. This analysis uses 23 different combinations of fingerprint types and sizes from RDKit, OpenEye, and Open Babel. The fingerprints were generated using randomly selected structures from ChEMBL 24.

Over the last five years deep learning has progressed tremendously in both image recognition and natural language processing. Now it is increasingly applied to other data rich fields. In drug discovery, recurrent neural networks (RNNs) have been shown to be an effective method to generate novel chemical structures in the form of SMILES. However, ligands generated by current methods have so far provided relatively low diversity and do not fully cover the whole chemical space occupied by known ligands. Here, we propose a new method (DrugEx) to discover de novo drug-like molecules. DrugEx is an RNN model (generator) trained through reinforcement learning which was integrated with a special exploration strategy. As a case study we applied our method to design ligands against the adenosine A2A receptor. From ChEMBL data, a machine learning model (predictor) was created to predict whether generated molecules are active or not. Based on this predictor as the reward function, the generator was trained by reinforcement learning without any further data. We then compared the performance of our method with two previously published methods, REINVENT and ORGANIC. We found that candidate molecules our model designed, and predicted to be active, had a larger chemical diversity, and better covered the chemical space of known ligands compared to the state-of-the-art.

The prediction of molecular targets is highly beneficial during the drug discovery process, be it for off-target elucidation or deconvolution of phenotypic screens. Here, we present OCEAN, a target prediction tool exclusively utilizing publically available ChEMBL data. OCEAN uses a heuristics approach based on a validation set containing almost 1000 drug ← → target relationships. New ChEMBL data (ChEMBL20 as well as ChEMBL21) released after the validation was used for a prospective OCEAN performance check. The success rates of OCEAN to predict correctly the targets within the TOP10 ranks are 77% for recently marketed drugs and 62% for all new ChEMBL20 compounds and 51% for all new ChEMBL21 compounds. OCEAN is also capable of identifying polypharmacological compounds; the success rate for molecules simultaneously hitting at least two targets is 64% to be correctly predicted within the TOP10 ranks. The source code of OCEAN can be found at http://www.github.com/rdkit/OCEAN.

An Internet portal accessible at www.gdb.unibe.ch has been set up to automatically generate color-coded similarity maps of the ChEMBL database in relation to up to two sets of active compounds taken from the enhanced Directory of Useful Decoys (eDUD), a random set of molecules, or up to two sets of user-defined reference molecules. These maps visualize the relationships between the selected compounds and ChEMBL in six different high dimensional chemical spaces, namely MQN (42-D molecular quantum numbers), SMIfp (34-D SMILES fingerprint), APfp (20-D shape fingerprint), Xfp (55-D pharmacophore fingerprint), Sfp (1024-bit substructure fingerprint), and ECfp4 (1024-bit extended connectivity fingerprint). The maps are supplied in form of Java based desktop applications called "similarity mapplets" allowing interactive content browsing and linked to a "Multifingerprint Browser for ChEMBL" (also accessible directly at www.gdb.unibe.ch ) to perform nearest neighbor searches. One can obtain six similarity mapplets of ChEMBL relative to random reference compounds, 606 similarity mapplets relative to single eDUD active sets, 30,300 similarity mapplets relative to pairs of eDUD active sets, and any number of similarity mapplets relative to user-defined reference sets to help visualize the structural diversity of compound series in drug optimization projects and their relationship to other known bioactive compounds.

Matched molecular pair analysis (MMPA) has become a major tool for analyzing large chemistry data sets for promising chemical transformations. However, the dependence of MMPA predictions on data constraints such as the number of pairs involved, experimental uncertainty, source of the experiments, and variability of the true physical effect has not yet been described. In this contribution the statistical basics for judging MMPA are analyzed. We illustrate the connection between overall MMPA statistics and individual pairs with a detailed comparison of average CHEMBL hERG MMPA results versus pairs with extreme transformation effects. Comparing the CHEMBL results to Novartis data, we find that significant transformation effects agree very well if the experimental uncertainty is considered. This indicates that caution must be exercised for predictions from insignificant MMPAs, yet highlights the robustness of statistically validated MMPA and shows that MMPA on public databases can yield results that are very useful for medicinal chemistry.

A matched molecular series is the general form of a matched molecular pair and refers to a set of two or more molecules with the same scaffold but different R groups at the same position. We describe Matsy, a knowledge-based method that uses matched series to predict R groups likely to improve activity given an observed activity order for some R groups. We compare the Matsy predictions based on activity data from ChEMBLdb to the recommendations of the Topliss tree and carry out a large scale retrospective test to measure performance. We show that the basis for predictive success is preferred orders in matched series and that this preference is stronger for longer series. The Matsy algorithm allows medicinal chemists to integrate activity trends from diverse medicinal chemistry programs and apply them to problems of interest as a Topliss-like recommendation or as a hypothesis generator to aid compound design.

While a plethora of ligands are known for the well studied α7 and α4β2 nicotinic acetylcholine receptor (nAChR), only very few ligands address the related α3β2 nAChR expressed in the central nervous system and at the neuromuscular junction. Starting with the public database ChEMBL organized in the chemical space of Molecular Quantum Numbers (MQN, a series of 42 integer value descriptors of molecular structure), a visual survey of nearest neighbors of the α7 nAChR partial agonist N-(3R)-1-azabicyclo[2.2.2]oct-3-yl-4-chlorobenzamide (PNU-282,987) pointed to N-(2-halobenzyl)-3-aminoquinuclidines as possible nAChR modulators. This simple "chemical space walk" was performed using a web-browser available at www.gdb.unibe.ch . Electrophysiological recordings revealed that these ligands represent a new and to date most potent class of positive allosteric modulators (PAMs) of the α3β2 nAChR, which also exert significant effects in vivo. The present discovery highlights the value of surveying chemical space neighbors of known drugs within public databases to uncover new pharmacology.

We carried out a systematic evaluation of target selectivity profiles across three recent large-scale biochemical assays of kinase inhibitors and further compared these standardized bioactivity assays with data reported in the widely used databases ChEMBL and STITCH. Our comparative evaluation revealed relative benefits and potential limitations among the bioactivity types, as well as pinpointed biases in the database curation processes. Ignoring such issues in data heterogeneity and representation may lead to biased modeling of drugs' polypharmacological effects as well as to unrealistic evaluation of computational strategies for the prediction of drug-target interaction networks. Toward making use of the complementary information captured by the various bioactivity types, including IC50, K(i), and K(d), we also introduce a model-based integration approach, termed KIBA, and demonstrate here how it can be used to classify kinase inhibitor targets and to pinpoint potential errors in database-reported drug-target interactions. An integrated drug-target bioactivity matrix across 52,498 chemical compounds and 467 kinase targets, including a total of 246,088 KIBA scores, has been made freely available.

The concept of data fusion - the combination of information from different sources describing the same object with the expectation to generate a more accurate representation - has found application in a very broad range of disciplines. In the context of ligand-based virtual screening (VS), data fusion has been applied to combine knowledge from either different active molecules or different fingerprints to improve similarity search performance. Machine-learning (ML) methods based on fusion of multiple homogeneous classifiers, in particular random forests, have also been widely applied in the ML literature. The heterogeneous version of classifier fusion - fusing the predictions from different model types - has been less explored. Here, we investigate heterogeneous classifier fusion for ligand-based VS using three different ML methods, RF, naïve Bayes (NB), and logistic regression (LR), with four 2D fingerprints, atom pairs, topological torsions, RDKit fingerprint, and circular fingerprint. The methods are compared using a previously developed benchmarking platform for 2D fingerprints which is extended to ML methods in this article. The original data sets are filtered for difficulty, and a new set of challenging data sets from ChEMBL is added. Data sets were also generated for a second use case: starting from a small set of related actives instead of diverse actives. The final fused model consistently outperforms the other approaches across the broad variety of targets studied, indicating that heterogeneous classifier fusion is a very promising approach for ligand-based VS. The new data sets together with the adapted source code for ML methods are provided in the Supporting Information .

Bioactivity databases are routinely used in drug discovery to look-up and, using prediction tools, to predict potential targets for small molecules. These databases are typically manually curated from patents and scientific articles. Apart from errors in the source document, the human factor can cause errors during the extraction process. These errors can lead to wrong decisions in the early drug discovery process. In the current work, we have compared bioactivity data from three large databases (ChEMBL, Liceptor, and WOMBAT) who have curated data from the same source documents. As a result, we are able to report error rate estimates for individual activity parameters and individual bioactivity databases. Small molecule structures have the greatest estimated error rate followed by target, activity value, and activity type. This order is also reflected in supplier-specific error rate estimates. The results are also useful in identifying data points for recuration. We hope the results will lead to a more widespread awareness among scientists on the frequencies and types of errors in bioactivity data.