Influence of search parameters and criteria on compound selection, promiscuity, and pan assay interference characteristics

Identification of New Toxicity Mechanisms in Drug-Induced Liver Injury through Systems Pharmacology

Among adverse drug reactions, drug-induced liver injury presents particular challenges because of its complexity, and the underlying mechanisms are still not completely characterized. Our knowledge of the topic is limited and based on the assumption that a drug acts on one molecular target. We have leveraged drug polypharmacology, i.e., the ability of a drug to bind multiple targets and thus perturb several biological processes, to develop a systems pharmacology platform that integrates all drug–target interactions. Our analysis sheds light on the molecular mechanisms of drugs involved in drug-induced liver injury and provides new hypotheses to study this phenomenon.

Probabilistic Random Forest improves bioactivity predictions close to the classification threshold by taking into account experimental uncertainty

Measurements of protein–ligand interactions have reproducibility limits due to experimental errors. Any model based on such assays will consequentially have such unavoidable errors influencing their performance which should ideally be factored into modelling and output predictions, such as the actual standard deviation of experimental measurements (σ) or the associated comparability of activity values between the aggregated heterogenous activity units (i.e., K_i versus IC₅₀ values) during dataset assimilation. However, experimental errors are usually a neglected aspect of model generation. In order to improve upon the current state-of-the-art, we herein present a novel approach toward predicting protein–ligand interactions using a Probabilistic Random Forest (PRF) classifier. The PRF algorithm was applied toward in silico protein target prediction across ~ 550 tasks from ChEMBL and PubChem. Predictions were evaluated by taking into account various scenarios of experimental standard deviations in both training and test sets and performance was assessed using fivefold stratified shuffled splits for validation. The largest benefit in incorporating the experimental deviation in PRF was observed for data points close to the binary threshold boundary, when such information was not considered in any way in the original RF algorithm. For example, in cases when σ ranged between 0.4–0.6 log units and when ideal probability estimates between 0.4–0.6, the PRF outperformed RF with a median absolute error margin of ~ 17%. In comparison, the baseline RF outperformed PRF for cases with high confidence to belong to the active class (far from the binary decision threshold), although the RF models gave errors smaller than the experimental uncertainty, which could indicate that they were overtrained and/or over-confident. Finally, the PRF models trained with putative inactives decreased the performance compared to PRF models without putative inactives and this could be because putative inactives were not assigned an experimental pXC₅₀ value, and therefore they were considered inactives with a low uncertainty (which in practice might not be true). In conclusion, PRF can be useful for target prediction models in particular for data where class boundaries overlap with the measurement uncertainty, and where a substantial part of the training data is located close to the classification threshold.

Interpretation of Compound Activity Predictions from Complex Machine Learning Models Using Local Approximations and Shapley Values

In qualitative or quantitative studies of structure-activity relationships (SARs), machine learning (ML) models are trained to recognize structural patterns that differentiate between active and inactive compounds. Understanding model decisions is challenging but of critical importance to guide compound design. Moreover, the interpretation of ML results provides an additional level of model validation based on expert knowledge. A number of complex ML approaches, especially deep learning (DL) architectures, have distinctive black-box character. Herein, a locally interpretable explanatory method termed Shapley additive explanations (SHAP) is introduced for rationalizing activity predictions of any ML algorithm, regardless of its complexity. Models resulting from random forest (RF), nonlinear support vector machine (SVM), and deep neural network (DNN) learning are interpreted, and structural patterns determining the predicted probability of activity are identified and mapped onto test compounds. The results indicate that SHAP has high potential for rationalizing predictions of complex ML models.

Large-Scale Comparison of Alternative Similarity Search Strategies with Varying Chemical Information Contents

Similarity searching (SS) is a core approach in computational compound screening and has a long tradition in pharmaceutical research. Over the years, different approaches have been introduced to increase the information content of search calculations and optimize the ability to detect compounds having similar activity. We present a large-scale comparison of distinct search strategies on more than 600 qualifying compound activity classes. Challenging test cases for SS were identified and used to evaluate different ways to further improve search performance, which provided a differentiated view of alternative search strategies and their relative performance. It was found that search results could not only be improved by increasing compound input information but also by focusing similarity calculations on database compounds. In the presence of multiple active reference compounds, asymmetric SS with high weights on chemical features of target compounds emerged as an overall preferred approach across many different activity classes. These findings have implications for practical virtual screening applications.

SuperDRUG2: a one stop resource for approved/marketed drugs

Regular monitoring of drug regulatory agency web sites and similar resources for information on new drug approvals and changes to legal status of marketed drugs is impractical. It requires navigation through several resources to find complete information about a drug as none of the publicly accessible drug databases provide all features essential to complement in silico drug discovery. Here, we propose SuperDRUG2 (http://cheminfo.charite.de/superdrug2) as a comprehensive knowledge-base of approved and marketed drugs. We provide the largest collection of drugs (containing 4587 active pharmaceutical ingredients) which include small molecules, biological products and other drugs. The database is intended to serve as a one-stop resource providing data on: chemical structures, regulatory details, indications, drug targets, side-effects, physicochemical properties, pharmacokinetics and drug–drug interactions. We provide a 3D-superposition feature that facilitates estimation of the fit of a drug in the active site of a target with a known ligand bound to it. Apart from multiple other search options, we introduced pharmacokinetics simulation as a unique feature that allows users to visualise the ‘plasma concentration versus time’ profile for a given dose of drug with few other adjustable parameters to simulate the kinetics in a healthy individual and poor or extensive metabolisers.

Exploration of Target Synergy in Cancer Treatment by Cell-Based Screening Assay and Network Propagation Analysis

Computational approaches have previously been introduced to predict compounds with activity against multiple targets or compound combinations with synergistic functional effects. By contrast, there are no computational studies available that explore combinations of targets that might act synergistically upon small molecule treatment. Herein, we introduce an approach designed to identify synergistic target pairs on the basis of cell-based screening data and compounds with known target annotations. The targets involved in forming synergistic pairs were analyzed through a novel network propagation algorithm for rationalizing possible common synergy mechanisms. This algorithm enabled further analysis of each synergistic target pair and the identification of "interactors", i.e., proteins with higher propagation scores than would be expected by adding the individual contributions of each target in the synergistic pair. We detected 137 synergistic target pairs including 51 unique targets. A global network analysis of these 51 targets made it possible to derive a subnetwork of proteins with significant synergy. Furthermore, interactors were identified for 87 synergistic target pairs upon individual analysis of the network propagation of each pair. These interactors were associated with pathways related to cancer and apoptosis, membrane transport, and steroid metabolism and provided possible explanations of synergistic effects.

SAR Matrix Method for Large-Scale Analysis of Compound Structure-Activity Relationships and Exploration of Multitarget Activity Spaces

As the number of compounds and the volume of bioactivity data rapidly grow, advanced computational methods are required to study structure-activity relationships (SARs) on a large scale. Herein, the SAR matrix (SARM) methodology is described that was designed to systematically extract structural relationships between bioactive compounds from large databases, explore structure-activity relationships, and navigate multitarget activity spaces, which is one of the core tasks in chemogenomics. In addition, the SARM approach was designed to visualize structural and structure-activity relationships, which is often of critical importance for making this information available in an intuitive form for practical applications.

The Catch-22 of Predicting hERG Blockade Using Publicly Accessible Bioactivity Data

Drug-induced inhibition of the human ether-à-go-go-related gene (hERG)-encoded potassium ion channels can lead to fatal cardiotoxicity. Several marketed drugs and promising drug candidates were recalled because of this concern. Diverse modeling methods ranging from molecular similarity assessment to quantitative structure-activity relationship analysis employing machine learning techniques have been applied to data sets of varying size and composition (number of blockers and nonblockers). In this study, we highlight the challenges involved in the development of a robust classifier for predicting the hERG end point using bioactivity data extracted from the public domain. To this end, three different modeling methods, nearest neighbors, random forests, and support vector machines, were employed to develop predictive models using different molecular descriptors, activity thresholds, and training set compositions. Our models demonstrated superior performance in external validations in comparison with those reported in the previous studies from which the data sets were extracted. The choice of descriptors had little influence on the model performance, with minor exceptions. The criteria used to filter bioactivity data, the activity threshold settings used to separate blockers from nonblockers, and the structural diversity of blockers in training data set were found to be the crucial indicators of model performance. Training sets based on a binary threshold of 1 μM/10 μM to separate blockers (IC₅₀/ K_i ≤ 1 μM) from nonblockers (IC₅₀/ K_i > 10 μM) provided superior performance in comparison with those defined using a single threshold (1 μM or 10 μM). A major limitation in using the public domain hERG activity data is the abundance of blockers in comparison with nonblockers at usual activity thresholds, since not many studies report the latter.

Drugs as habitable planets in the space of dark chemical matter

A recent study demonstrated antifungal activity of dark chemical matter (DCM) compounds that were otherwise inactive in more than 100 HTS assays. These compounds were proposed to possess unique activity and 'clean' safety profiles. Here, we present an outlook of the promiscuity and safety of these compounds by retrospectively comparing their chemical and biological spaces with those of drugs. Significant amounts of marketed drugs (16%), withdrawn drugs (16.5%) and natural compounds (3.5%) share structural identity with DCM. Compound promiscuity assessment indicates that dark matter compounds could potentially interact with multiple biological targets. Further, thousands of DCM compounds showed presence of frequent-hitting pan-assay interference compound (PAINS) substructures. In light of these observations, filtering these compounds from screening libraries can be an irrevocable loss.

Do Fragments and Crystallization Additives Bind Similarly to Drug-like Ligands?

The success of fragment-based drug design (FBDD) hinges upon the optimization of low-molecular-weight compounds (MW < 300 Da) with weak binding affinities to lead compounds with high affinity and selectivity. Usually, structural information from fragment-protein complexes is used to develop ideas about the binding mode of similar but drug-like molecules. In this regard, crystallization additives such as cryoprotectants or buffer components, which are highly abundant in crystal structures, are frequently ignored. Thus, the aim of this study was to investigate the information present in protein complexes with fragments as well as those with additives and how they relate to the binding modes of their drug-like counterparts. We present a thorough analysis of the binding modes of crystallographic additives, fragments, and drug-like ligands bound to four diverse targets of wide interest in drug discovery and highly represented in the Protein Data Bank: cyclin-dependent kinase 2, β-secretase 1, carbonic anhydrase 2, and trypsin. We identified a total of 630 unique molecules bound to the catalytic binding sites, among them 31 additives, 222 fragments, and 377 drug-like ligands. In general, we observed that, independent of the target, protein-fragment interaction patterns are highly similar to those of drug-like ligands and mostly cover the residues crucial for binding. Crystallographic additives are also able to show conserved binding modes and recover the residues important for binding in some of the cases. Moreover, we show evidence that the information from fragments and drug-like ligands can be applied to rescore docking poses in order to improve the prediction of binding modes.

Promiscuity of inhibitors of human protein kinases at varying data confidence levels and test frequencies

Kinase inhibitors from screening data. Shown are two analogs that were tested against 10 (left) and 13 (right) different kinases. The inhibitor on the left was active against a single kinase and the one on the right against three kinases.

Identification and analysis of promiscuity cliffs formed by bioactive compounds and experimental implications

For three promiscuity cliffs (enclosed), cliff compounds, their promiscuity degrees (PDs), and color-coded substitution sites are shown. Comparison of these cliffs suggests the design of a new analog to further explore promiscuity.

Analyzing compound activity records and promiscuity degrees in light of publication statistics

For the generation of contemporary databases of bioactive compounds, activity information is usually extracted from the scientific literature. However, when activity data are analyzed, source publications are typically no longer taken into consideration. Therefore, compound activity data selected from ChEMBL were traced back to thousands of original publications, activity records including compound, assay, and target information were systematically generated, and their distributions across the literature were determined. In addition, publications were categorized on the basis of activity records. Furthermore, compound promiscuity, defined as the ability of small molecules to specifically interact with multiple target proteins, was analyzed in light of publication statistics, thus adding another layer of information to promiscuity assessment. It was shown that the degree of compound promiscuity was not influenced by increasing numbers of source publications. Rather, most non-promiscuous as well as promiscuous compounds, regardless of their degree of promiscuity, originated from single publications, which emerged as a characteristic feature of the medicinal chemistry literature.

Activity-relevant similarity values for fingerprints and implications for similarity searching

A largely unsolved problem in chemoinformatics is the issue of how calculated compound similarity relates to activity similarity, which is central to many applications. In general, activity relationships are predicted from calculated similarity values. However, there is no solid scientific foundation to bridge between calculated molecular and observed activity similarity. Accordingly, the success rate of identifying new active compounds by similarity searching is limited. Although various attempts have been made to establish relationships between calculated fingerprint similarity values and biological activities, none of these has yielded generally applicable rules for similarity searching. In this study, we have addressed the question of molecular versus activity similarity in a more fundamental way. First, we have evaluated if activity-relevant similarity value ranges could in principle be identified for standard fingerprints and distinguished from similarity resulting from random compound comparisons. Then, we have analyzed if activity-relevant similarity values could be used to guide typical similarity search calculations aiming to identify active compounds in databases. It was found that activity-relevant similarity values can be identified as a characteristic feature of fingerprints. However, it was also shown that such values cannot be reliably used as thresholds for practical similarity search calculations. In addition, the analysis presented herein helped to rationalize differences in fingerprint search performance.

Activity-relevant similarity values for fingerprints and implications for similarity searching

A largely unsolved problem in chemoinformatics is the issue of how calculated compound similarity relates to activity similarity, which is central to many applications. In general, activity relationships are predicted from calculated similarity values. However, there is no solid scientific foundation to bridge between calculated molecular and observed activity similarity. Accordingly, the success rate of identifying new active compounds by similarity searching is limited. Although various attempts have been made to establish relationships between calculated fingerprint similarity values and biological activities, none of these has yielded generally applicable rules for similarity searching. In this study, we have addressed the question of molecular versus activity similarity in a more fundamental way. First, we have evaluated if activity-relevant similarity value ranges could in principle be identified for standard fingerprints and distinguished from similarity resulting from random compound comparisons. Then, we have analyzed if activity-relevant similarity values could be used to guide typical similarity search calculations aiming to identify active compounds in databases. It was found that activity-relevant similarity values can be identified as a characteristic feature of fingerprints. However, it was also shown that such values cannot be reliably used as thresholds for practical similarity search calculations. In addition, the analysis presented herein helped to rationalize differences in fingerprint search performance.

Systematic assessment of analog relationships between bioactive compounds and promiscuity of analog sets

Compound–analog relationships. Shown is an active compound with four substitution sites, two of which are explored with four and seven different R-groups, respectively.

WITHDRAWN--a resource for withdrawn and discontinued drugs

Post-marketing drug withdrawals can be associated with various events, ranging from safety issues such as reported deaths or severe side-effects, to a multitude of non-safety problems including lack of efficacy, manufacturing, regulatory or business issues. During the last century, the majority of drugs voluntarily withdrawn from the market or prohibited by regulatory agencies was reported to be related to adverse drug reactions. Understanding the underlying mechanisms of toxicity is of utmost importance for current and future drug discovery. Here, we present WITHDRAWN, a resource for withdrawn and discontinued drugs publicly accessible at http://cheminfo.charite.de/withdrawn. Today, the database comprises 578 withdrawn or discontinued drugs, their structures, important physico-chemical properties, protein targets and relevant signaling pathways. A special focus of the database lies on the drugs withdrawn due to adverse reactions and toxic effects. For approximately one half of the drugs in the database, safety issues were identified as the main reason for withdrawal. Withdrawal reasons were extracted from the literature and manually classified into toxicity types representing adverse effects on different organs. A special feature of the database is the presence of multiple search options which will allow systematic analyses of withdrawn drugs and their mechanisms of toxicity.

Systematic Assessment of Molecular Selectivity at the Level of Targets, Bioactive Compounds, and Structural Analogues

Through systematic mining of compound activity data, the target selectivity of bioactive compounds was systematically explored. The analysis was facilitated by applying, extending, and combining the concepts of target cliffs, selectivity cliffs, and matched molecular pairs. Selectivity relationships were explored at different levels including targets, individual bioactive compounds, and pairs of structural analogues. A variety of targets were identified for which active compounds were consistently nonselective or, by contrast, exclusively selective, making it possible to prioritize, or de-prioritize, targets for compound development. Furthermore, many chemical modifications were detected that altered compound selectivity in a well-defined manner including small structural changes that converted nonselective into target-selective compounds or inverted the target selectivity of active compounds. A large knowledge base of selectivity relationships across pharmaceutical targets and chemical modifications that alter selectivity was generated; this has been made freely available to the scientific community as a part of this investigation.

Strong nonadditivity as a key structure-activity relationship feature: distinguishing structural changes from assay artifacts

Nonadditivity in protein–ligand affinity data represents highly instructive structure–activity relationship (SAR) features that indicate structural changes and have the potential to guide rational drug design. At the same time, nonadditivity is a challenge for both basic SAR analysis as well as many ligand-based data analysis techniques such as Free-Wilson Analysis and Matched Molecular Pair analysis, since linear substituent contribution models inherently assume additivity and thus do not work in such cases. While structural causes for nonadditivity have been analyzed anecdotally, no systematic approaches to interpret and use nonadditivity prospectively have been developed yet. In this contribution, we lay the statistical framework for systematic analysis of nonadditivity in a SAR series. First, we develop a general metric to quantify nonadditivity. Then, we demonstrate the non-negligible impact of experimental uncertainty that creates apparent nonadditivity, and we introduce techniques to handle experimental uncertainty. Finally, we analyze public SAR data sets for strong nonadditivity and use recourse to the original publications and available X-ray structures to find structural explanations for the nonadditivity observed. We find that all cases of strong nonadditivity (ΔΔpK_i and ΔΔpIC₅₀ > 2.0 log units) with sufficient structural information to generate reasonable hypothesis involve changes in binding mode. With the appropriate statistical basis, nonadditivity analysis offers a variety of new attempts for various areas in computer-aided drug design, including the validation of scoring functions and free energy perturbation approaches, binding pocket classification, and novel features in SAR analysis tools.

Structural and Activity Profile Relationships Between Drug Scaffolds

Core structures of current drugs have been assembled and their structural relationships and activity profiles have been explored. Drug scaffolds were frequently involved in different types of structural relationships. In addition, a variety of activity profile relationships between structurally related drug scaffolds were detected, ranging from closely overlapping to distinct profiles. Furthermore, when structural and activity profile relationships of scaffolds from drugs and bioactive compounds were compared, systematic differences were detected. Consensus activity profiles were introduced as a new approach for the qualitative and quantitative assessment of activity similarity of structurally related drugs represented by the same scaffold. On the basis of consensus activity profiles, scaffolds representing drugs active against distinct targets can be distinguished from drugs having similar target profiles and target hypotheses can be derived for individual drugs. Given the results of our analysis, drug scaffolds have been systematically organized according to structural and activity profile criteria. Our scaffold sets and the associated information are made freely available.

Identification and analysis of the currently available high-confidence three-dimensional activity cliffs

Shown is an exemplary 3D-cliff formed by two crystallographic ligands with highly similar binding modes and a significant difference in potency. The site of a major interaction difference between these compounds is encircled.