Structure-based predictions of activity cliffs

Anatomy of Potency Predictions Focusing on Structural Analogues with Increasing Potency Differences Including Activity Cliffs

Potency predictions are popular in compound design and optimization but are complicated by intrinsic limitations. Moreover, even for nonlinear methods, activity cliffs (ACs, formed by structural analogues with large potency differences) represent challenging test cases for compound potency predictions. We have devised a new test system for potency predictions, including AC compounds, that is based on partitioned matched molecular pairs (MMP) and makes it possible to monitor prediction accuracy at the level of analogue pairs with increasing potency differences. The results of systematic predictions using different machine learning and control methods on MMP-based data sets revealed increasing prediction errors when potency differences between corresponding training and test compounds increased, including large prediction errors for AC compounds. At the global level, these prediction errors were not apparent due to the statistical dominance of analogue pairs with small potency differences. Test compounds from such pairs were accurately predicted and determined the observed global prediction accuracy. Shapley value analysis, an explainable artificial intelligence approach, was applied to identify structural features determining potency predictions using different methods. The analysis revealed that numerical predictions of different regression models were determined by features that were shared by MMP partner compounds or absent in these compounds, with opposing effects. These findings provided another rationale for accurate predictions of similar potency values for structural analogues and failures in predicting the potency of AC compounds.

Exploring QSAR models for activity-cliff prediction

INTRODUCTION AND METHODOLOGY

Pairs of similar compounds that only differ by a small structural modification but exhibit a large difference in their binding affinity for a given target are known as activity cliffs (ACs). It has been hypothesised that QSAR models struggle to predict ACs and that ACs thus form a major source of prediction error. However, the AC-prediction power of modern QSAR methods and its quantitative relationship to general QSAR-prediction performance is still underexplored. We systematically construct nine distinct QSAR models by combining three molecular representation methods (extended-connectivity fingerprints, physicochemical-descriptor vectors and graph isomorphism networks) with three regression techniques (random forests, k-nearest neighbours and multilayer perceptrons); we then use each resulting model to classify pairs of similar compounds as ACs or non-ACs and to predict the activities of individual molecules in three case studies: dopamine receptor D2, factor Xa, and SARS-CoV-2 main protease.

RESULTS AND CONCLUSIONS

Our results provide strong support for the hypothesis that indeed QSAR models frequently fail to predict ACs. We observe low AC-sensitivity amongst the evaluated models when the activities of both compounds are unknown, but a substantial increase in AC-sensitivity when the actual activity of one of the compounds is given. Graph isomorphism features are found to be competitive with or superior to classical molecular representations for AC-classification and can thus be employed as baseline AC-prediction models or simple compound-optimisation tools. For general QSAR-prediction, however, extended-connectivity fingerprints still consistently deliver the best performance amongs the tested input representations. A potential future pathway to improve QSAR-modelling performance might be the development of techniques to increase AC-sensitivity.

Large-scale prediction of activity cliffs using machine and deep learning methods of increasing complexity

Activity cliffs (AC) are formed by pairs of structural analogues that are active against the same target but have a large difference in potency. While much of our knowledge about ACs has originated from the analysis and comparison of compounds and activity data, several studies have reported AC predictions over the past decade. Different from typical compound classification tasks, AC predictions must be carried out at the level of compound pairs representing ACs or nonACs. Most AC predictions reported so far have focused on individual methods or comparisons of two or three approaches and only investigated a few compound activity classes (from 2 to 10). Although promising prediction accuracy has been reported in most cases, different system set-ups, AC definitions, methods, and calculation conditions were used, precluding direct comparisons of these studies. Therefore, we have carried out a large-scale AC prediction campaign across 100 activity classes comparing machine learning methods of greatly varying complexity, ranging from pair-based nearest neighbor classifiers and decision tree or kernel methods to deep neural networks. The results of our systematic predictions revealed the level of accuracy that can be expected for AC predictions across many different compound classes. In addition, prediction accuracy did not scale with methodological complexity but was significantly influenced by memorization of compounds shared by different ACs or nonACs. In many instances, limited training data were sufficient for building accurate models using different methods and there was no detectable advantage of deep learning over simpler approaches for AC prediction. On a global scale, support vector machine models performed best, by only small margins compared to others including simple nearest neighbor classifiers.

Development and evaluation of a java-based deep neural network method for drug response predictions

Accurate prediction of drug response is a crucial step in personalized medicine. Recently, deep learning techniques have been witnessed with significant breakthroughs in a variety of areas including biomedical research and chemogenomic applications. This motivated us to develop a novel deep learning platform to accurately and reliably predict the response of cancer cells to different drug treatments. In the present work, we describe a Java-based implementation of deep neural network method, termed JavaDL, to predict cancer responses to drugs solely based on their chemical features. To this end, we devised a novel cost function and added a regularization term which suppresses overfitting. We also adopted an early stopping strategy to further reduce overfit and improve the accuracy and robustness of our models. To evaluate our method, we compared with several popular machine learning and deep neural network programs and observed that JavaDL either outperformed those methods in model building or obtained comparable predictions. Finally, JavaDL was employed to predict drug responses of several aggressive breast cancer cell lines, and the results showed robust and accurate predictions with r ² as high as 0.81.

Exposing the Limitations of Molecular Machine Learning with Activity Cliffs

Machine learning has become a crucial tool in drug discovery and chemistry at large, e.g., to predict molecular properties, such as bioactivity, with high accuracy. However, activity cliffs─pairs of molecules that are highly similar in their structure but exhibit large differences in potency─have received limited attention for their effect on model performance. Not only are these edge cases informative for molecule discovery and optimization but also models that are well equipped to accurately predict the potency of activity cliffs have increased potential for prospective applications. Our work aims to fill the current knowledge gap on best-practice machine learning methods in the presence of activity cliffs. We benchmarked a total of 24 machine and deep learning approaches on curated bioactivity data from 30 macromolecular targets for their performance on activity cliff compounds. While all methods struggled in the presence of activity cliffs, machine learning approaches based on molecular descriptors outperformed more complex deep learning methods. Our findings highlight large case-by-case differences in performance, advocating for (a) the inclusion of dedicated "activity-cliff-centered" metrics during model development and evaluation and (b) the development of novel algorithms to better predict the properties of activity cliffs. To this end, the methods, metrics, and results of this study have been encapsulated into an open-access benchmarking platform named MoleculeACE (Activity Cliff Estimation, available on GitHub at: https://github.com/molML/MoleculeACE). MoleculeACE is designed to steer the community toward addressing the pressing but overlooked limitation of molecular machine learning models posed by activity cliffs.

Pharmacophore Modeling of Targets Infested with Activity Cliffs via Molecular Dynamics Simulation Coupled with QSAR and Comparison with other Pharmacophore Generation Methods: KDR as Case Study

Activity cliffs (ACs) are defined as pairs of structurally similar compounds with large difference in their potencies against certain biotarget. We recently proposed that potent AC members induce significant entropically-driven conformational modifications of the target that unveil additional binding interactions, while their weakly-potent counterparts are enthalpically-driven binders with little influence on the protein target. We herein propose to extract pharmacophores for ACs-infested target(s) from molecular dynamics (MD) frames of purely "enthalpic" potent binder(s) complexed within the particular target. Genetic function algorithm/machine learning (GFA/ML) can then be employed to search for the best possible combination of MD pharmacophore(s) capable of explaining bioactivity variations within a list of inhibitors. We compared the performance of this approach with established ligand-based and structure-based methods. Kinase inserts domain receptor (KDR) was used as a case study. KDR plays a crucial role in angiogenic signaling and its inhibitors have been approved in cancer treatment. Interestingly, GFA/ML selected, MD-based, pharmacophores were of comparable performances to ligand-based and structure-based pharmacophores. The resulting pharmacophores and QSAR models were used to capture hits from the national cancer institute list of compounds. The most active hit showed anti-KDR IC50 of 2.76 µM.

ACGCN: Graph Convolutional Networks for Activity Cliff Prediction between Matched Molecular Pairs

One of the interesting issues in drug-target interaction studies is the activity cliff (AC), which is usually defined as structurally similar compounds with large differences in activity toward a common target. The AC is of great interest in medicinal chemistry as it may provide clues to understanding the complex properties of the target proteins, paving the way for practical applications aimed at the discovery of more potent drugs. In this paper, we propose graph convolutional networks for the prediction of AC and designate the proposed models as Activity Cliff prediction using Graph Convolutional Networks (ACGCNs). The results show that ACGCNs outperform several off-the-shelf methods when predicting ACs of three popular target data sets for thrombin, Mu opioid receptor, and melanocortin receptor. Finally, we utilize gradient-weighted class activation mapping to visualize activation weights at nodes in the molecular graphs, demonstrating its potential to contribute to the ability to identify important substructures for molecular docking.

Discovery and SAR Evolution of Pyrazole Azabicyclo[3.2.1]octane Sulfonamides as a Novel Class of Non-Covalent N-Acylethanolamine-Hydrolyzing Acid Amidase (NAAA) Inhibitors for Oral Administration

Inhibition of intracellular N-acylethanolamine-hydrolyzing acid amidase (NAAA) activity is a promising approach to manage the inflammatory response under disabling conditions. In fact, NAAA inhibition preserves endogenous palmitoylethanolamide (PEA) from degradation, thus increasing and prolonging its anti-inflammatory and analgesic efficacy at the inflamed site. In the present work, we report the identification of a potent, systemically available, novel class of NAAA inhibitors, featuring a pyrazole azabicyclo[3.2.1]octane structural core. After an initial screening campaign, a careful structure-activity relationship study led to the discovery of endo-ethoxymethyl-pyrazinyloxy-8-azabicyclo[3.2.1]octane-pyrazole sulfonamide 50 (ARN19689), which was found to inhibit human NAAA in the low nanomolar range (IC50 = 0.042 μM) with a non-covalent mechanism of action. In light of its favorable biochemical, in vitro and in vivo drug-like profile, sulfonamide 50 could be regarded as a promising pharmacological tool to be further investigated in the field of inflammatory conditions.

Advances in exploring activity cliffs

The activity cliff (AC) concept is of comparable relevance for medicinal chemistry and chemoinformatics. An AC is defined as a pair of structurally similar compounds with a large potency difference against a given target. In medicinal chemistry, ACs are of interest because they reveal small chemical changes with large potency effects, a concept referred to as structure-activity relationship (SAR) discontinuity. Computationally, ACs can be systematically identified, going far beyond individual compound series considered during lead optimization. Large-scale analysis of ACs has revealed characteristic features across many different compound activity classes. The way in which the molecular similarity and potency difference criteria have been addressed for defining ACs distinguishes between different generations of ACs and mirrors the evolution of the AC concept. We discuss different stages of this evolutionary path and highlight recent advances in AC research.

Quantum Mechanical Investigation of Three-Dimensional Activity Cliffs Using the Molecules-in-Molecules Fragmentation-Based Method

The concept of activity cliff (AC) (i.e., a small structural modification resulting in a substantial bioactivity change) is widely encountered in medicinal chemistry during compound design. Whereas the study of ACs is of high interest as it provides a wealth of opportunities for effective drug design, its practical application in the actual drug development process has been difficult because of significant computational challenges. To provide some understanding of the ACs, we have carried out a rigorous quantum-mechanical investigation of the electronic interactions of a wide range of ACs (205 cliffs formed by 261 protein-ligand complexes covering 37 different receptor types) using multilayer molecules-in-molecules (MIM) fragmentation-based methodology. The MIM methodology enables performing accurate high-level quantum mechanical (QM) calculations at a substantially lower computational cost, while allowing for a quantitative decomposition of the protein-ligand binding energy into the contributions from individual residues, solvation, and entropy. Our investigation in this study is mainly focused on whether the QM binding energy calculation can correctly identify the higher potency cliff partner for a given ligand pair having a sufficiently high activity difference. We have also analyzed the effect of including crystal water molecules as a part of the receptor as well as the impact of ligand desolvation energy on the correct identification of the more potent ligand in a cliff pair. Our analysis reveals that, in the majority of the cases, the AC prediction could be significantly improved by carefully identifying the critical crystal water molecules, whereas the contribution from the ligand desolvation also remains essential. Additionally, we have exploited the residue-specific interaction energies provided by MIM to identify the key residues and interaction hot-spots that are responsible for the experimentally observed drastic activity changes. The results show that our MIM fragmentation-based protocol provides comprehensive interaction energy profiles that can be employed to understand the distinctiveness of ligand modifications, for potential applications in structure-based drug design.

Pharmacophore modeling of JAK1: A target infested with activity-cliffs

Janus kinase 1 (JAK1) is protein kinase involved in autoimmune diseases (AIDs). JAK1 inhibitors have shown promising results in treating AIDs. JAK1 inhibitors are known to exhibit regions of SAR discontinuity or activity cliffs (ACs). ACs represent fundamental challenge to successful QSAR/pharmacophore modeling because QSAR modeling rely on the basic premise that activity is a smooth continuous function of structure. We propose that ACs exist because active ACs members exhibit subtle, albeit critical, enthalpic features absent from their inactive twins. In this context we compared the performances of two computational modeling workflows in extracting valid pharmacophores from 151 diverse JAK1 inhibitors that include ACs: QSAR-guided pharmacophore selection versus docking-based comparative intermolecular contacts analysis (db-CICA). The two methods were judged based on the receiver operating characteristic (ROC) curves of their corresponding pharmacophore models and their abilities to distinguish active members among established JAK1 ACs. db-CICA modeling significantly outperformed ligand-based pharmacophore modeling. The resulting optimal db-CICA pharmacophore was used as virtual search query to scan the National Cancer Institute (NCI) database for novel JAK1 inhibitory leads. The most active hit showed IC₅₀ of 1.04 μM. This study proposes the use of db-CICA modeling as means to extract valid pharmacophores from SAR data infested with ACs.

Introducing a new category of activity cliffs combining different compound similarity criteria

Activity cliffs (ACs) are pairs of structurally similar or analogous active compounds with large differences in potency against the same target. For identifying and analyzing ACs, similarity and potency difference criteria must be determined and consistently applied. This can be done in various ways, leading to different types of ACs. In this work, we introduce a new category of ACs by combining different similarity criteria, including the formation of matched molecular pairs and structural isomer relationships. A systematic computational search identified such ACs in compounds with activity against a variety of targets. In addition to other ACs exclusively formed by structural isomers, the newly introduced category of ACs is rich in structure-activity relationship (SAR) information, straightforward to interpret from a chemical perspective, and further extends the current spectrum of ACs.

Evolving Concept of Activity Cliffs

Activity cliffs (ACs) are generally defined as pairs or groups of structurally similar compounds that are active against the same target but have large differences in potency. Accordingly, ACs capture chemical modifications that strongly influence biological activity. Therefore, they are of particular interest in structure-activity relationship (SAR) analysis and compound optimization. The AC concept is much more complex than it may appear at a first glance, especially if one aims to represent ACs computationally and identify them systematically. To these ends, molecular similarity and potency difference criteria must be carefully considered for AC assessment. Furthermore, ACs are often perceived differently in medicinal and computational chemistry, depending on whether they are studied on a case-by-case basis or systematically. For practical applications, intuitive access to AC information plays a major role. Over the years, the AC concept has been further refined and extended. Herein, we review the evolution of the AC concept, emphasizing new analysis schemes and findings that help to better understand ACs and extract SAR knowledge from them.

Experimental study and computational modelling of cruzain cysteine protease inhibition by dipeptidyl nitriles

Chagas disease affects millions of people in Latin America. This disease is caused by the protozoan parasite Trypanossoma cruzi. The cysteine protease cruzain is a key enzyme for the survival and propagation of this parasite lifecycle. Nitrile-based inhibitors are efficient inhibitors of cruzain that bind by forming a covalent bond with this enzyme. Here, three nitrile-based inhibitors dubbed Neq0409, Neq0410 and Neq0570 were synthesized, and the thermodynamic profile of the bimolecular interaction with cruzain was determined using isothermal titration calorimetry (ITC). The result suggests the inhibition process is enthalpy driven, with a detrimental contribution of entropy. In addition, we have used hybrid Quantum Mechanical/Molecular Mechanical (QM/MM) and Molecular Dynamics (MD) simulations to investigate the reaction mechanism of reversible covalent modification of cruzain by Neq0409, Neq0410 and Neq0570. The computed free energy profile shows that the nucleophilic attack of Cys25 on the carbon C1 of inhibitiors and the proton transfer from His162 to N1 of the dipeptidyl nitrile inhibitor take place in a single step. The calculated free energy of the inhibiton reaction is in agreement with covalent experimental binding. Altogether, the results reported here suggests that nitrile-based inhibitors are good candidates for the development of reversible covalent inhibitors of cruzain and other cysteine proteases.

Nuclear Receptors Database Including Negative Data (NR-DBIND): A Database Dedicated to Nuclear Receptors Binding Data Including Negative Data and Pharmacological Profile

Nuclear receptors (NRs) are transcription factors that regulate gene expression in various physiological processes through their interactions with small hydrophobic molecules. They constitute an important class of targets for drugs and endocrine disruptors and are widely studied for both health and environment concerns. Since the integration of negative data can be critical for accurate modeling of ligand activity profiles, we manually collected and annotated NRs interaction data (positive and negative) through a sharp review of the corresponding literature. 15 116 positive and negative interactions data are provided for 28 NRs together with 593 PDB structures in the freely available Nuclear Receptors Database Including Negative Data ( http://nr-dbind.drugdesign.fr ). The NR-DBIND contains the most extensive information about interaction data on NRs, which should bring valuable information to chemists, biologists, pharmacologists and toxicologists.

Structure-Based Kinase Profiling To Understand the Polypharmacological Behavior of Therapeutic Molecules

Several drugs elicit their therapeutic efficacy by modulating multiple cellular targets and possess varied polypharmacological actions. The identification of the molecular targets of a potent bioactive molecule is essential in determining its overall polypharmacological profile. Experimental procedures are expensive and time-consuming. Therefore, computational approaches are actively implemented in rational drug discovery. Here, we demonstrate a computational pipeline, based on reverse virtual screening technique using several consensus scoring strategies, and perform structure-based kinase profiling of 12 FDA-approved drugs. This target prediction showed an overall good performance, with an average AU-ROC greater than 0.85 for most drugs, and identified the true targets even at the top 2% cutoff. In contrast, 10 non-kinase binder drugs exhibited lower binding efficiency and appeared in the bottom of ranking list. Subsequently, we validated this pipeline on a potent therapeutic molecule, mahanine, whose polypharmacological profile related to targeting kinases is unknown. Our target-prediction method identified different kinases. Furthermore, we have experimentally validated that mahanine is able to modulate multiple kinases that are involved in cross-talk with different signaling molecules, which thereby exhibits its polypharmacological action. More importantly, in vitro kinase assay exhibited the inhibitory effect of mahanine on two such predicted kinases' (mTOR and VEGFR2) activity, with IC₅₀ values being ∼12 and ∼22 μM, respectively. Next, we generated a comprehensive drug-protein interaction fingerprint that explained the basis of their target selectivity. We observed that it is controlled by variations in kinase conformations followed by significant differences in crucial hydrogen-bond and van der Waals interactions. Such structure-based kinase profiling could provide useful information in revealing the unknown targets of therapeutic molecules from their polypharmacological behavior and would assist in drug discovery.

Ligand-biased ensemble receptor docking (LigBEnD): a hybrid ligand/receptor structure-based approach

Ligand docking to flexible protein molecules can be efficiently carried out through ensemble docking to multiple protein conformations, either from experimental X-ray structures or from in silico simulations. The success of ensemble docking often requires the careful selection of complementary protein conformations, through docking and scoring of known co-crystallized ligands. False positives, in which a ligand in a wrong pose achieves a better docking score than that of native pose, arise as additional protein conformations are added. In the current study, we developed a new ligand-biased ensemble receptor docking method and composite scoring function which combine the use of ligand-based atomic property field (APF) method with receptor structure-based docking. This method helps us to correctly dock 30 out of 36 ligands presented by the D3R docking challenge. For the six mis-docked ligands, the cognate receptor structures prove to be too different from the 40 available experimental Pocketome conformations used for docking and could be identified only by receptor sampling beyond experimentally explored conformational subspace.

Getting SMARt in drug discovery: chemoinformatics approaches for mining structure–multiple activity relationships

In light of the high relevance of polypharmacology, multi-target screening is a major trend in drug discovery.

Matched Molecular Pair Analysis in Short: Algorithms, Applications and Limitations

Molecular matched pair (MMP) analysis has been used for more than 40 years within molecular design and is still an important tool to analyse potency data and other compound properties. The methods used to find matched pairs range from manual inspection, through supervised methods to unsupervised methods, which are able to find previously unknown molecular pairs. Recent publications demonstrate the value of automatic MMP analysis of publicly available bioactivity databases. The MMP concept has its limitations, but because of its easy to use and intuitive nature, it will remain one of the most important tools in the toolbox of many drug designers.

De Novo Design at the Edge of Chaos

Computational medicinal chemistry offers viable strategies for finding, characterizing, and optimizing innovative pharmacologically active compounds. Technological advances in both computer hardware and software as well as biological chemistry have enabled a renaissance of computer-assisted "de novo" design of molecules with desired pharmacological properties. Here, we present our current perspective on the concept of automated molecule generation by highlighting chemocentric methods that may capture druglike chemical space, consider ligand promiscuity for hit and lead finding, and provide fresh ideas for the rational design of customized screening of compound libraries.

Historeceptomic Fingerprints for Drug-Like Compounds

Most drugs exert their beneficial and adverse effects through their combined action on several different molecular targets (polypharmacology). The true molecular fingerprint of the direct action of a drug has two components: the ensemble of all the receptors upon which a drug acts and their level of expression in organs/tissues. Conversely, the fingerprint of the adverse effects of a drug may derive from its action in bystander tissues. The ensemble of targets is almost always only partially known. Here we describe an approach improving upon and integrating both components: in silico identification of a more comprehensive ensemble of targets for any drug weighted by the expression of those receptors in relevant tissues. Our system combines more than 300,000 experimentally determined bioactivity values from the ChEMBL database and 4.2 billion molecular docking scores. We integrated these scores with gene expression data for human receptors across a panel of human tissues to produce drug-specific tissue-receptor (historeceptomics) scores. A statistical model was designed to identify significant scores, which define an improved fingerprint representing the unique activity of any drug. These multi-dimensional historeceptomic fingerprints describe, in a novel, intuitive, and easy to interpret style, the holistic, in vivo picture of the mechanism of any drug's action. Valuable applications in drug discovery and personalized medicine, including the identification of molecular signatures for drugs with polypharmacologic modes of action, detection of tissue-specific adverse effects of drugs, matching molecular signatures of a disease to drugs, target identification for bioactive compounds with unknown receptors, and hypothesis generation for drug/compound phenotypes may be enabled by this approach. The system has been deployed at drugable.org for access through a user-friendly web site.

Integration of Ligand and Structure Based Approaches for CSAR-2014

The prediction of binding poses and affinities is an area of active interest in computer-aided drug design (CADD). Given the documented limitations with either ligand or structure based approaches, we employed an integrated approach and developed a rapid protocol for binding mode and affinity predictions. This workflow was applied to the three protein targets of Community Structure-Activity Resource-2014 (CSAR-2014) exercise: Factor Xa (FXa), Spleen Tyrosine Kinase (SYK), and tRNA (guanine-N(1))-methyltransferase (TrmD). Our docking and scoring workflow incorporates compound clustering and ligand and protein structure based pharmacophore modeling, followed by local docking, minimization, and scoring. While the former part of the protocol ensures high-quality ligand alignments and mapping, the subsequent minimization and scoring provides the predicted binding modes and affinities. We made blind predictions of docking pose for 1, 5, and 14 ligands docked into 1, 2, and 12 crystal structures of FXa, SYK, and TrmD, respectively. The resulting 174 poses were compared with cocrystallized structures (1, 5, and 14 complexes) made available at the end of CSAR. Our predicted poses were related to the experimentally determined structures with a mean root-mean-square deviation value of 3.4 Å. Further, we were able to classify high and low affinity ligands with the area under the curve values of 0.47, 0.60, and 0.69 for FXa, SYK, and TrmD, respectively, indicating the validity of our approach in at least two of the three systems. Detailed critical analysis of the results and CSAR methodology ranking procedures suggested that a straightforward application of our workflow has limitations, as some of the performance measures do not reflect the actual utility of pose and affinity predictions in the biological context of individual systems.

Activity landscape sweeping: insights into the mechanism of inhibition and optimization of DNMT1 inhibitors

Inhibitors of DNA methyltransferases have distinct structure–activity relationships as revealed by the activity landscape sweeping study discussed in this work.