Classical QSAR Research Articles

Machine learning accelerates pharmacophore-based virtual screening of MAO inhibitors

Nowadays, an efficient and robust virtual screening procedure is crucial in the drug discovery process, especially when performed on large and chemically diverse databases. Virtual screening methods, like molecular docking and classic QSAR models, are limited in their ability to handle vast numbers of compounds and to learn from scarce data, respectively. In this study, we introduce a universal methodology that uses a machine learning-based approach to predict docking scores without the need for time-consuming molecular docking procedures. The developed protocol yielded 1000 times faster binding energy predictions than classical docking-based screening. The proposed predictive model learns from docking results, allowing users to choose their preferred docking software without relying on insufficient and incoherent experimental activity data. The methodology described employs multiple types of molecular fingerprints and descriptors to construct an ensemble model that further reduces prediction errors and is capable of delivering highly precise docking score values for monoamine oxidase ligands, enabling faster identification of promising compounds. An extensive pharmacophore-constrained screening of the ZINC database resulted in a selection of 24 compounds that were synthesized and evaluated for their biological activity. A preliminary screen discovered weak inhibitors of MAO-A with a percentage efficiency index close to a known drug at the lowest tested concentration. The approach presented here can be successfully applied to other biological targets as target-specific knowledge is not incorporated at the screening phase.

QSAR Modeling Based on Conformation Ensembles Using a Multi-Instance Learning Approach

Modern QSAR approaches have wide practical applications in drug discovery for designing potentially bioactive molecules. If such models are based on the use of 2D descriptors, important information contained in the spatial structures of molecules is lost. The major problem in constructing models using 3D descriptors is the choice of a putative bioactive conformation, which affects the predictive performance. The multi-instance (MI) learning approach considering multiple conformations in model training could be a reasonable solution to the above problem. In this study, we implemented several multi-instance algorithms, both conventional and based on deep learning, and investigated their performance. We compared the performance of MI-QSAR models with those based on the classical single-instance QSAR (SI-QSAR) approach in which each molecule is encoded by either 2D descriptors computed for the corresponding molecular graph or 3D descriptors issued for a single lowest energy conformation. The calculations were carried out on 175 data sets extracted from the ChEMBL23 database. It is demonstrated that (i) MI-QSAR outperforms SI-QSAR in numerous cases and (ii) MI algorithms can automatically identify plausible bioactive conformations.

Machine Learning augmented docking studies of aminothioureas at the SARS-CoV-2-ACE2 interface.

The current pandemic outbreak clearly indicated the urgent need for tools allowing fast predictions of bioactivity of a large number of compounds, either available or at least synthesizable. In the computational chemistry toolbox, several such tools are available, with the main ones being docking and structure-activity relationship modeling either by classical linear QSAR or Machine Learning techniques. In this contribution, we focus on the comparison of the results obtained using different docking protocols on the example of the search for bioactivity of compounds containing N-N-C(S)-N scaffold at the S-protein of SARS-CoV-2 virus with ACE2 human receptor interface. Based on over 1800 structures in the training set we have predicted binding properties of the complete set of nearly 600000 structures from the same class using the Machine Learning Random Forest Regressor approach.

Identification and in silico prediction of metabolites of tebufenozide derivatives by major human cytochrome P450 isoforms

Cytochrome P450 (CYP) enzymes constitute a superfamily of heme-containing monooxygenases. CYPs are involved in the metabolism of many chemicals such as drugs and agrochemicals. Therefore, examining the metabolic reactions by each CYP isoform is important to elucidate their substrate recognition mechanisms. The clarification of these mechanisms may be useful not only for the development of new drugs and agrochemicals, but also for risk assessment of chemicals. In our previous study, we identified the metabolites of tebufenozide, an insect growth regulator, formed by two human CYP isoforms: CYP3A4 and CYP2C19. The accessibility of each site of tebufenozide to the reaction center of CYP enzymes and the susceptibility of each hydrogen atom for metabolism by CYP enzymes were evaluated by a docking simulation and hydrogen atom abstraction energy estimation at the density functional theory level, respectively. In this study, the same in silico prediction method was applied to the metabolites of tebufenozide derivatives by major human CYPs (CYP1A2, 2C9, 2C19, 2D6, and 3A4). In addition, the production rate of the metabolites by CYP3A4 was quantitively analyzed by frequency based on docking simulation and hydrogen atom abstraction energy using the classical QSAR approach. Then, the obtained QSAR model was applied to predict the sites of metabolism and the metabolite production order by each CYP isoform.

Chemical in vitro bioactivity profiles are not informative about the long-term in vivo endocrine mediated toxicity

Endocrine disrupting chemicals raise a lot of interest and concern regarding their risk for human health and the environment. They represent a broad variety of natural and synthetic chemicals with different levels of endocrine activity evaluation. In particular, for high production volume chemicals, new methods are required to enable the evaluation of the vast number of chemicals for their potential to alter the endocrine system and prioritize them for deeper characterization. The ToxCast program from the US EPA provides data from high throughput screening assays to develop computational tools aimed at rapid in vitro bioactivity screening and prioritization.Using publicly available data (ToxCast and ToxRef databases), we evaluate whether in vitro assay evaluations could predict in vivo outcomes observed in rat long-term studies for more than 400 chemicals. We focus on effects observed in three endocrine and two sex accessory organs and 42 in vitro assays related to pathways associated with endocrine related toxicity.First, using simple statistical correlation we demonstrate that there is no mutual linear correlation between the selected in vitro assays and any in vivo outcome, with balanced accuracies around 50% for each assay-outcome pair. Then, by applying machine learning to investigate potential non-linear correlations, we show that the combination of different in vitro assays is not correlated with the long-term in vivo effects and cannot help to predict them since balanced accuracies are also around 50%. Moreover, the prediction based on in vitro assays is not better than the one based on classical QSAR methods. This study highlights that the selected in vitro assays do not provide information about in vivo outcomes observed in endocrine and associated organs in long-term rat in vivo studies and stresses the need for the development of in vitro assays that reflect the compounds’ pharmacokinetic properties.

Docking, Interaction Fingerprint, and Three-Dimensional Quantitative Structure-Activity Relationship (3D-QSAR) of Sigma1 Receptor Ligands, Analogs of the Neuroprotective Agent RC-33.

The human Sigma1 receptor (S1R), which has been identified as a target with an important role in neuropsychological disorders, was first crystallized 3 years ago. Since S1R structure has no relation with another previous crystallized structures, the presence of the new crystal is an important hallmark for the design of agonists and antagonists against this important target. Some years ago, our group identified RC-33, a potent and selective S1R agonist, endowed with neuroprotective properties. In this work, drawing on new structural information, we studied the interactions of RC-33 and its analogs with the S1R binding site by using computational methods such as docking, interaction fingerprints, and receptor-guided alignment three dimensional quantitative structure–activity relationship (3D-QSAR). We found that RC-33 and its analogs adopted similar orientations within S1R binding site, with high similitude with orientations of the crystallized ligands; such information was used for identifying the residues involved in chemical interactions with ligands. Furthermore, the structure-activity relationship of the studied ligands was adequately described considering classical QSAR tests. All relevant aspects of the interactions between the studied compounds and S1R were covered here, through descriptions of orientations, binding interactions, and features that influence differential affinities. In this sense, the present results could be useful in the future design of novel S1R modulators.

MaNGA: a novel multi-niche multi-objective genetic algorithm for QSAR modelling.

Quantitative structure-activity relationship (QSAR) modelling is currently used in multiple fields to relate structural properties of compounds to their biological activities. This technique is also used for drug design purposes with the aim of predicting parameters that determine drug behaviour. To this end, a sophisticated process, involving various analytical steps concatenated in series, is employed to identify and fine-tune the optimal set of predictors from a large dataset of molecular descriptors (MDs). The search of the optimal model requires to optimize multiple objectives at the same time, as the aim is to obtain the minimal set of features that maximizes the goodness of fit and the applicability domain (AD). Hence, a multi-objective optimization strategy, improving multiple parameters in parallel, can be applied. Here we propose a new multi-niche multi-objective genetic algorithm that simultaneously enables stable feature selection as well as obtaining robust and validated regression models with maximized AD. We benchmarked our method on two simulated datasets. Moreover, we analyzed an aquatic acute toxicity dataset and compared the performances of single- and multi-objective fitness functions on different regression models. Our results show that our multi-objective algorithm is a valid alternative to classical QSAR modelling strategy, for continuous response values, since it automatically finds the model with the best compromise between statistical robustness, predictive performance, widest AD, and the smallest number of MDs. The python implementation of MaNGA is available at https://github.com/Greco-Lab/MaNGA. Supplementary data are available at Bioinformatics online.

Insight into structural features of phenyltetrazole derivatives as ABCG2 inhibitors for the treatment of multidrug resistance in cancer

ABSTRACT ABCG2 is the principal ABC transporter involved in the multidrug resistance of breast cancer. Looking at the current demand in the development of ABCG2 inhibitors for the treatment of multidrug-resistant cancer, we have explored structural requirements of phenyltetrazole derivatives for ABCG2 inhibition by combining classical QSAR, Bayesian classification modelling and molecular docking studies. For classical QSAR, structural descriptors were calculated from the free software tool PaDEL-descriptor. Stepwise multiple linear regression (SMLR) was used for model generation. A statistically significant model was generated and validated with different parameters (For training set: r = 0.825; Q 2 = 0.570 and for test set: r = 0.894, r2 pred = 0.783). The predicted model was found to satisfy the Golbraikh and Trospha criteria for model acceptability. Bayesian classification modelling was also performed (ROC scores were 0.722 and 0.767 for the training and test sets, respectively). Finally, the binding interactions of phenyltetrazole type inhibitor with the ABCG2 receptor were mapped with the help of molecular docking study. The result of the docking analysis is aligned with the classical QSAR and Bayesian classification studies. The combined modelling study will guide the medicinal chemists to act faster in the drug discovery of ABCG2 inhibitors for the management of resistant breast cancer.

Stacked Generalization with Applicability Domain Outperforms Simple QSAR on in Vitro Toxicological Data.

The development of in silico tools able to predict bioactivity and toxicity of chemical substances is a powerful solution envisioned to assess toxicity as early as possible. To enable the development of such tools, the ToxCast program has generated and made publicly available in vitro bioactivity data for thousands of compounds. The goal of the present study is to characterize and explore the data from ToxCast in terms of Machine Learning capability. For this, a large scale analysis on the entire database has been performed to build models to predict bioactivities measured in in vitro assays. Simple classical QSAR algorithms (ANN, SVM, LDA, random forest, and Bayesian) were first applied on the data, and the results of these algorithms suggested that they do not seem to be well-suited for data sets with a high proportion of inactive compounds. The study then showed for the first time that the use of an ensemble method named "Stacked generalization" could improve the model performance on this type of data. Indeed, for 61% of 483 models, the Stacked method led to models with higher performance. Moreover, the combination of this ensemble method with an applicability domain filter allows one to assess the reliability of the predictions for further compound prioritization. In particular we showed that for 50% of the models, the ROC score is better if we do not consider the compounds that are not within the applicability domain.

Clef topo-toxicity by cubic representations of organophosphates

The present article studies a series of organophosphate analog molecules used as pesticides. The study involves calculations of some of the most relevant connectivity indices (as Wiener Index, Topological Efficiency Index, Novel Topological Efficiency Index) for the two-dimensional graphs of the studied molecules obtained through the classic QSAR rules. A cubic system with nodes, conceived for packing/superposing the vertices of the studied graphs led to different adjacencies of the vertices in the packed graphs, so to new connectivity indices, named three-dimensional connectivity indices. The new connectivity indices prove better correlation properties with aspects as toxicity and biological activity as enzyme inhibitors (expressed as lethal doses or inhibitory concentrations) than the two-dimensional indices thus the cubic mode of the topological study of molecular graphs could represent a variant of initial in silico studies for determining in analog series the molecules with best activities or improved toxicological characteristics.

Classical QSAR and Docking Simulation of 4-Pyridone Derivatives for Their Antimalarial Activity.

In this work, the minimum energy structures of 22 4-pyridone derivatives have been optimized at Density Functional Theory level, and several quantum molecular, including electronic and thermodynamic descriptors, were computed for these substrates in order to obtain a statistical and meaningful QSAR equation. In this sense, by using multiple linear regressions, five mathematical models have been obtained. The best model with only four descriptors (r2 = 0.86, Q2 = 0.92, S.E.P = 0.38) was validated by the leave-one-out cross-validation method. The antimalarial activity can be explained by the combination of the four mentioned descriptors e.g., electronic potential, dipolar momentum, partition coefficient and molar refractivity. The statistical parameters of this model suggest that it is robust enough to predict the antimalarial activity of new possible compounds; consequently, three small chemical modifications into the structural core of these compounds were performed specifically on the most active compound of the series (compound 13). These three new suggested compounds were leveled as 13A, 13B and 13C, and the predicted biological antimalarial activity is 0.02 µM, 0.03 µM, and 0.07 µM, respectively. In order to complement these results focused on the possible action mechanism of the substrates, a docking simulation was included for these new structures as well as for the compound 13 and the docking scores (binding affinity) obtained for the interaction of these substrates with the cytochrome bc1, were −7.5, −7.2, −6.9 and −7.5 kcal/mol for 13A, 13B, 13C and compound 13, respectively, which suggests that these compounds are good candidates for its biological application in this illness.

Molecular Highlights about Leishmanicidal Activity and Cytotoxicity of Quinoxaline Derivatives: A Combination of Ligand-based QSAR/QSTR Approaches

Background: Leishmaniasis is a neglected tropical disease, considered endemic in almost 98 countries worldwide, with 250 million new cases estimated annually. The usual drugs are considered toxic, expensive and frequently ineffective, justifying the search for new lead compounds. We presented here ligand-based classical QSAR study involving the leishmanicidal activity of quinoxaline derivatives and molecular descriptors of diverse nature, aimed to understand their molecular activities from a quantitative point of view. Additionally, the cytotoxic activity of the compounds was also modeled in a QSTR approach, in order to help the design of more active and safe leishmanicidal compounds. Methods: Geometry optimizations were performed by semiempirical and DFT quantum-chemical methods. Molecular descriptors were calculated by E-Dragon online platform and submitted to a variable selection procedure, using Ordered Predictors Selector algorithm. The best QSAR and QSTR models were selected by MLR modeling and structure-activity relationships were considered, in accordance to the successful of the validation parameters. Results: For leishmanicidal activity, five descriptors were selected as the most important, providing a robust and predictive QSAR model via Multiple Linear Regression (MLR) (R2 = 0.71, Qloo 2 = 0.60, Qext 2 = 0.76 and rtest = 0.87). On the other hand, cytotoxicity was better related to seven descriptors in QSTR approach (R2 = 0.77, Qloo 2 = 0.65, Qext 2 = 0.85 and rtest = 0.92). Analysis of descriptors influences on both activities revealed that there are common features that could be explored to provide a most leishmanicidal and less cytotoxic compound, at the same time, in order to design more active and safe compounds. Conclusion: In establishing of the structure-activity relationships and interpreting the physical sense of the descriptors, we can conclude that molecules containing sulfoxyl, sulfonyl and halogens tend to be most active, but molecules with large volumes and electronegative atoms tend to be less active. In addition, in a very interesting way, we could observe that small molecules containing sulfoxyl, sulfonyl and halogens tend to be less cytotoxic and larger molecules with fewer atoms of greater mass and electronegativity than carbon tend to be more cytotoxic. Fortunately, the combination of these characteristics can be explored in the synthesis and evaluation of new safe derivatives with better leishmanicidal profile.

SBDDとclassical QSAR 解析によるチミジンホスホリラーゼ阻害薬の分子設計

SBDDとclassical QSAR 解析によるチミジンホスホリラーゼ阻害薬の分子設計

The Application of Classical QSAR to Agrochemical Research

Since the first paper for the QSAR procedure was published in 1962, a number of agrochemical QSAR studies have been accumulated. Among them, some studies were to construct models to design new compounds of the improved activity profiles. Others were to search for a clue for molecular mechanisms of action and/or physicochemical backgrounds for some “compulsory or favorable” substructures and substituents within the compound series. In this article, some examples of studies from these two categories are reviewed.

Advantages and limitations of classic and 3D QSAR approaches in nano-QSAR studies based on biological activity of fullerene derivatives.

In this contribution, the advantages and limitations of two computational techniques that can be used for the investigation of nanoparticles activity and toxicity: classic nano-QSAR (Quantitative Structure–Activity Relationships employed for nanomaterials) and 3D nano-QSAR (three-dimensional Quantitative Structure–Activity Relationships, such us Comparative Molecular Field Analysis, CoMFA/Comparative Molecular Similarity Indices Analysis, CoMSIA analysis employed for nanomaterials) have been briefly summarized. Both approaches were compared according to the selected criteria, including: efficiency, type of experimental data, class of nanomaterials, time required for calculations and computational cost, difficulties in the interpretation. Taking into account the advantages and limitations of each method, we provide the recommendations for nano-QSAR modellers and QSAR model users to be able to determine a proper and efficient methodology to investigate biological activity of nanoparticles in order to describe the underlying interactions in the most reliable and useful manner.Electronic supplementary materialThe online version of this article (doi:10.1007/s11051-016-3564-1) contains supplementary material, which is available to authorized users.

Structure–activity relationships of dibenzoylhydrazines for the inhibition of P-glycoprotein-mediated quinidine transport

We previously demonstrated that dibenzoylhydrazines (DBHs) are not only P-glycoprotein (P-gp) substrates, but also inhibitors. In the present study, we evaluated the inhibition of P-gp-mediated quinidine transport by two series of DBHs and performed a classical QSAR analysis and docking simulation in order to investigate the mechanisms underlying P-gp substrate/inhibitor recognition. The results of the QSAR analysis identified the hydrophobic factor as the most important for inhibitory activities, while electronic and steric effects also influenced the activities. The different substituent effects observed in each series suggested the different binding modes of each series of DBHs, which was supported by the results of the docking simulation.

Quantitative structure–hydrophobicity relationships of molecular fragments and beyond

Quantitative structure-property relationship (QSPR) models were firstly established for the hydrophobic substituent constant (πX) using the theoretical descriptors derived solely from electrostatic potentials (EPSs) at the substituent atoms. The descriptors introduced are found to be related to hydrogen-bond basicity, hydrogen-bond acidity, cavity, or dipolarity/polarizability terms in linear solvation energy relationship, which endows the models good interpretability. The predictive capabilities of the models constructed were also verified by rigorous Monte Carlo cross-validation. Then, eight groups of meta- or para-disubstituted benzenes and one group of substituted pyridines were investigated. QSPR models for individual systems were achieved with the ESP-derived descriptors. Additionally, two QSPR models were also established for Rekker's fragment constants (foct), which is a secondary-treatment quantity and reflects average contribution of the fragment to logP. It has been demonstrated that the descriptors derived from ESPs at the fragments, can be well used to quantitatively express the relationship between fragment structures and their hydrophobic properties, regardless of the attached parent structure or the valence state. Finally, the relations of Hammett σ constant and ESP quantities were explored. It implies that σ and π, which are essential in classic QSAR and represent different type of contributions to biological activities, are also complementary in interaction site.

Computational chemogenomics: is it more than inductive transfer?

High-throughput assays challenge us to extract knowledge from multi-ligand, multi-target activity data. In QSAR, weights are statically fitted to each ligand descriptor with respect to a single endpoint or target. However, computational chemogenomics (CG) has demonstrated benefits of learning from entire grids of data at once, rather than building target-specific QSARs. A possible reason for this is the emergence of inductive knowledge transfer (IT) between targets, providing statistical robustness to the model, with no assumption about the structure of the targets. Relevant protein descriptors in CG should allow one to learn how to dynamically adjust ligand attribute weights with respect to protein structure. Hence, models built through explicit learning (EL) by including protein information, while benefitting from IT enhancement, should provide additional predictive capability, notably for protein deorphanization. This interplay between IT and EL in CG modeling is not sufficiently studied. While IT is likely to occur irrespective of the injected target information, it is not clear whether and when boosting due to EL may occur. EL is only possible if protein description is appropriate to the target set under investigation. The key issue here is the search for evidence of genuine EL exceeding expectations based on pure IT. We explore the problem in the context of Support Vector Regression, using more than 9,400 pKi values of 31 GPCRs, where compound-protein interactions are represented by the concatenation of vectorial descriptions of compounds and proteins. This provides a unified framework to generate both IT-enhanced and potentially EL-enabled models, where the difference is toggled by supplied protein information. For EL-enabled models, protein information includes genuine protein descriptors such as typical sequence-based terms, but also the experimentally determined affinity cross-correlation fingerprints. These latter benchmark the expected behavior of a quasi-ideal descriptor capturing the actual functional protein-protein relatedness, and therefore thought to be the most likely to enable EL. EL- and IT-based methods were benchmarked alongside classical QSAR, with respect to cross-validation and deorphanization challenges. A rational method for projecting benchmarked methodologies into a strategy space is given, in the aims that the projection will provide directions for the types of molecule designs possible using a given methodology. While EL-enabled strategies outperform classical QSARs and favorably compare to similar published results, they are, in all respects evaluated herein, not strongly distinguished from IT-enhanced models. Moreover, EL-enabled strategies failed to prove superior in deorphanization challenges. Therefore, this paper raises caution that, contrary to common belief and intuitive expectation, the benefits of chemogenomics models over classical QSAR are quite possibly due less to the injection of protein-related information, and rather impacted more by the effect of inductive transfer, due to simultaneous learning from all of the modeled endpoints. These results show that the field of protein descriptor research needs further improvements to truly realize the expected benefit of EL.

Limits to molecular matched-pair analysis: the experimental uncertainty case.

Matched-Molecular Pair (MMP) analysis has recently emerged as a data analysis technique in medicinal chemistry. It quickly gained scientific momentum because it tackles key questions in lead optimization. In contrast to classical global QSAR models that attempt to predict the absolute numbers of ADME (absorption, distribution, metabolism, excretion) and toxicological properties, MMP analyses predict the difference in (bio-) chemical properties that can be expected due to small chemical modifications to lead structures, with a much smaller and well-controlled error than global QSAR models. The power of MMP analysis depends on the number of previously documented similar molecular transformations, whereas the definition of chemical similarity plays a key role: the more generous the definition of similarity of the anchoring region, the more examples are available. The more strict the definition of similarity, the lower the variability and thus the clearer the effect on ADME-Tox parameters, but also the less data pairs will be available [1]. The (bio-) chemical effect and the significance of the results depends on the experimental uncertainty (=noise) in the data. There is a clear mathematical association between the noise level and the minimum activity difference necessary for statistical significance. Here we demonstrate how the experimental uncertainty and variability[2,3] affect Matched Molecular Pair Analysis. It can be shown that for small sample sizes (Context-specific MMPs), the activity differences have to be very large in order to be statistically significant. A full equation for the estimation of minimum significant activity difference, depending on the number of samples, standard deviation of the measurements and the true variance of the biochemical effect is developed. The influence of consistency of assay setups can directly be quantified via the variability and practical consequences for MMP analysis will be presented.

Optimal descriptor as a translator of eclectic information into the prediction of membrane damage by means of various TiO2 nanoparticles

The increasing use of nanomaterials incorporated into consumer products leads to the need for developing approaches to establish “quantitative structure–activity relationships” (QSARs) for various nanomaterials. However, the molecular structure as rule is not available for nanomaterials at least in its classic meaning. An possible alternative of classic QSAR (based on the molecular structure) is the using of data on physicochemical features of TiO2 nanoparticles. The damage to cellular membranes (unitsL−1) by means of various TiO2 nanoparticles is examined as the endpoint.