On some aspects of validation of predictive quantitative structure-activity relationship models

Deep Learning in Drug Discovery

Artificial neural networks had their first heyday in molecular informatics and drug discovery approximately two decades ago. Currently, we are witnessing renewed interest in adapting advanced neural network architectures for pharmaceutical research by borrowing from the field of "deep learning". Compared with some of the other life sciences, their application in drug discovery is still limited. Here, we provide an overview of this emerging field of molecular informatics, present the basic concepts of prominent deep learning methods and offer motivation to explore these techniques for their usefulness in computer-assisted drug discovery and design. We specifically emphasize deep neural networks, restricted Boltzmann machine networks and convolutional networks.

Predicting carcinogenicity of organic compounds based on CPDB

Cancer is a major killer of human health and predictions for the carcinogenicity of chemicals are of great importance. In this article, predictive models for the carcinogenicity of organic compounds using QSAR methods for rats and mice were developed based on the data from CPDB. The models was developed based on the data of specific target site liver and classified according to sex of rats and mice. Meanwhile, models were also classified according to whether there is a ring in the molecular structure in order to reduce the diversity of molecular structure. Therefore, eight local models were developed in the final. Taking into account the complexity of carcinogenesis and in order to obtain as much information, DRAGON descriptors were selected as the variables used to develop models. Fitting ability, robustness and predictive power of the models were assessed according to the OECD principles. The external predictive coefficients for validation sets of each model were in the range of 0.711-0.906, and for the whole data in each model were all greater than 0.8, which represents that all models have good predictivity. In order to study the mechanism of carcinogenesis, standardized regression coefficients were calculated for all predictor variables. In addition, the effect of animal sex on carcinogenesis was compared and a trend that female showed stronger tolerance for cancerogen than male in both species was appeared.

Understanding the quantitative structure–activity relationship of acetylcholinesterase inhibitors for the treatment of Alzheimer's disease

Alzheimer's disease (AD) is the most common cause of dementia in old aged people and clinically used drugs for treatment are associated with side effects. Thus, there is a current demand for the discovery and development of new potential molecules. However, the recent advances in drug therapy have challenged the predominance of the disease. In this manuscript, an attempt has been made to develop the 2D and 3D quantitative structure–activity relationship (QSAR) models for a series of rutaecarpine, quinazolines and 7,8-dehydrorutaecarpine derivatives to obtain insights to Acetylcholinesterase (AChE) inhibition. Five different QSAR models have been generated and validated using a set of 52 compounds comprising of varying scaffolds with IC50 values ranging from 11,000 nM to 0.6 nM. These AChE-specific prediction models (M1–M5) adequately reflect the structure–activity relationship of the existing AChE inhibitors. Out of all developed models, QSAR model generated using ADME properties has been found to be the best with satisfactory statistical significance (regression (r2) of 0.9309 and regression adjusted coefficient of variation [Formula: see text] of 0.9194). The QSAR models highlight the importance of aromatic moiety as their presence in the structure influence the biological activity. Additional insights on the compounds show that acyclic amines attached to side chain have lower activity than cyclic amines. The QSAR models pinpointing structural basis for the AChEIs suggest new guidelines for the design of novel molecules.

Monte Carlo QSAR models for predicting organophosphate inhibition of acetycholinesterase

A series of 278 organophosphate compounds acting as acetylcholinesterase inhibitors has been studied. The Monte Carlo method was used as a tool for building up one-variable quantitative structure-activity relationship (QSAR) models for acetylcholinesterase inhibition activity based on the principle that the target endpoint is treated as a random event. As an activity, bimolecular rate constants were used. The QSAR models were based on optimal descriptors obtained from Simplified Molecular Input-Line Entry System (SMILES) used for the representation of molecular structure. Two modelling approaches were examined: (1) 'classic' training-test system where the QSAR model was built with one random split into a training, test and validation set; and (2) the correlation balance based QSAR models were built with two random splits into a sub-training, calibration, test and validation set. The DModX method was used for defining the applicability domain. The obtained results suggest that studied activity can be determined with the application of QSAR models calculated with the Monte Carlo method since the statistical quality of all build models was very good. Finally, structural indicators for the increase and the decrease of the bimolecular rate constant are defined. The possibility of using these results for the computer-aided design of new organophosphate compounds is presented.

QSAR models for thiophene and imidazopyridine derivatives inhibitors of the Polo-Like Kinase 1

The inhibitory activity of 103 thiophene and 33 imidazopyridine derivatives against Polo-Like Kinase 1 (PLK1) expressed as pIC50 (-logIC50) was predicted by QSAR modeling. Multivariate linear regression (MLR) was employed to model the relationship between 0D and 3D molecular descriptors and biological activities of molecules using the replacement method (MR) as variable selection tool. The 136 compounds were separated into several training and test sets. Two splitting approaches, distribution of biological data and structural diversity, and the statistical experimental design procedure D-optimal distance were applied to the dataset. The significance of the training set models was confirmed by statistically higher values of the internal leave one out cross-validated coefficient of determination (Q2) and external predictive coefficient of determination for the test set (Rtest2). The model developed from a training set, obtained with the D-optimal distance protocol and using 3D descriptor space along with activity values, separated chemical features that allowed to distinguish high and low pIC50 values reasonably well. Then, we verified that such model was sufficient to reliably and accurately predict the activity of external diverse structures. The model robustness was properly characterized by means of standard procedures and their applicability domain (AD) was analyzed by leverage method.

The rm2 metrics and regression through origin approach: reliable and useful validation tools for predictive QSAR models (Commentary on 'Is regression through origin useful in external validation of QSAR models?')

Quantitative structure-activity relationship (QSAR) is an in silico technique which can be used in drug discovery, environmental fate modeling, property and toxicity prediction of chemical entities and regulatory toxicology. The predictive potential of a QSAR model is judged from various validation metrics in order to evaluate how well it is capable to predict endpoint values of new untested compounds. The rm2 group of metrics is one of the stringent validation metrics currently used by the QSAR fraternity in different reports. We scrutinized a recently published paper which raised an issue that the constructed criteria based on regression through origin (RTO) are not optimal and there is a significant difference in the rm2 metrics values computed from different statistical software packages. According to our point of view, the conclusion drawn in this paper appears to be misleading. Any inconsistency in the software algorithms has nothing to do with the calculation of rm2 metrics, as such computation is not limited by the use of any specific software, rather it depends only on fundamental mathematical formulae that are well established. However, it is a concern to the QSAR users that Excel and SPSS can return different results for the metrics using the RTO method. Thus, a proper validation of the software tool is required before its use for computation of any validation metric.

Is regression through origin useful in external validation of QSAR models?

The external validation of QSAR models is crucial to ensure their reliability for assessing new chemicals. The most widely used criteria for external validations, which has been applied in hundreds of more recent QSAR studies are the Golbraikh-Tropsha and Roy methods which these criteria are based on the regression through origin (RTO). In this study, the calculations of the deviation parameters such as absolute errors are used for ascertaining the difference between training and test sets to evaluate the prediction capability of the models. However, these results were not in a good agreement with the proposed criteria for external validation and there is an inconsistency in the definition and calculation of r(2) of RTO and therefore the constructed criteria based on RTO is not optimal. Instead, the calculation of model errors for training and test sets and compare them, provide a possible reliable method to external validation of QSAR models.

Machine learning estimates of natural product conformational energies

Machine learning has been used for estimation of potential energy surfaces to speed up molecular dynamics simulations of small systems. We demonstrate that this approach is feasible for significantly larger, structurally complex molecules, taking the natural product Archazolid A, a potent inhibitor of vacuolar-type ATPase, from the myxobacterium Archangium gephyra as an example. Our model estimates energies of new conformations by exploiting information from previous calculations via Gaussian process regression. Predictive variance is used to assess whether a conformation is in the interpolation region, allowing a controlled trade-off between prediction accuracy and computational speed-up. For energies of relaxed conformations at the density functional level of theory (implicit solvent, DFT/BLYP-disp3/def2-TZVP), mean absolute errors of less than 1 kcal/mol were achieved. The study demonstrates that predictive machine learning models can be developed for structurally complex, pharmaceutically relevant compounds, potentially enabling considerable speed-ups in simulations of larger molecular structures.

Molecular dynamics simulations provide insight into the dynamic behavior of molecules, e.g., into the adopted spatial arrangements of its atoms over time. Methods differ in the approximations they employ, resulting in a trade-off between accuracy and speed that ranges from highly accurate but expensive quantum mechanical calculations to fast but more inaccurate molecular mechanics force fields. Machine learning, a sub-discipline of artificial intelligence, provides algorithms that learn from data, that is, make predictions based on previously seen examples. By starting with a few expensive quantum mechanical calculations, training a machine learning algorithm on them, and then using the resulting model to carry out the molecular dynamics simulation, one can improve the accuracy/speed trade-off. We have developed and applied such a hybrid quantum mechanics/machine learning approach to Archazolid A, a natural product from the myxobacterium Archangium gephyra and a potent inhibitor of vacuolar-type ATPase. By dynamically refining our model over the course of the simulation, we achieve errors of less than 1 kcal/mol while saving over 40% of the quantum mechanical calculations. Our study demonstrates the feasibility of predictive machine learning models for the dynamics of structurally complex, pharmaceutically relevant compounds, potentially enabling considerable speed-ups in simulations of even larger biomolecular structures.

Development and validation of regression-based QSAR models for quantification of contributions of molecular fragments to skin sensitization potency of diverse organic chemicals

In our present work, we have developed regression-based QSAR models for skin sensitization potential of 51 diverse organic chemicals. The objective behind the present work is to determine the influence of different molecular features on the skin sensitizing potential of chemicals. Among several models developed, the two best ones are discussed to unveil specific information regarding the contribution of different structural and physicochemical features towards the property of skin sensitization. The QSAR models suggested that aromatic compounds are more skin sensitizing than aliphatic ones, but in the case of carbonyl compounds, aliphatic ones are more skin sensitizing than aromatic ones. Other descriptors such as LUMO and <2-Atype_H-47> signify the importance of the electrophilic and hydrophilic character respectively of the molecules for showing skin sensitizing property. Another two descriptors, <Dipole-mag-2.72> and (3)χC also exert significant contributions towards the skin sensitization potential of the chemicals. Further, it is observed that the nitrogen atoms (nN), triple bonds (nTB) and also the fragment Al-C(=X)-Al (Atype_C38) are responsible for skin sensitizing property. All the above information provides additional support for further research involving identification of the skin sensitization potential of new chemicals.

Preliminary studies on model development for rodent toxicity and its interspecies correlation with aquatic toxicities of pharmaceuticals

Environmental toxicity due to pharmaceuticals has been an issue of serious concern for long time. Development of chemometric models with reliable predictive power has been considered as an effective tool for the design of new drug agents with reduced or without ecotoxic potential. Considering a higher degree of similarity in genetic homology towards drug receptor with mammals, we have used a dataset of 194 compounds with reported rodent, fish, daphnia and algae toxicity data for extrapolation of their toxicity towards humans. Allowing for rodents as the most surrogate to human physiology, attempts have also been made to develop interspecies correlation models keeping rodent toxicity as dependent variable so that any drug without reported rodent toxicity can be predicted using fish, daphnia or algae toxicity data which can be consequently extrapolated to human toxicity. The developed models have been subjected to multiple validation strategies. Acceptable results have been obtained in both cases of direct and interspecies extrapolation quantitative structure-activity relationship models.

Materiomics: an -omics approach to biomaterials research

The past fifty years have seen a surge in the use of materials for clinical application, but in order to understand and exploit their full potential, the scientific complexity at both sides of the interface--the material on the one hand and the living organism on the other hand--needs to be considered. Technologies such as combinatorial chemistry, recombinant DNA as well as computational multi-scale methods can generate libraries with a very large number of material properties whereas on the other side, the body will respond to them depending on the biological context. Typically, biological systems are investigated using both holistic and reductionist approaches such as whole genome expression profiling, systems biology and high throughput genetic or compound screening, as already seen, for example, in pharmacology and genetics. The field of biomaterials research is only beginning to develop and adopt these approaches, an effort which we refer to as "materiomics". In this review, we describe the current status of the field, and its past and future impact on the biomedical sciences. We outline how materiomics sets the stage for a transformative change in the approach to biomaterials research to enable the design of tailored and functional materials for a variety of properties in fields as diverse as tissue engineering, disease diagnosis and de novo materials design, by combining powerful computational modelling and screening with advanced experimental techniques.

Molecular docking and receptor-specific 3D-QSAR studies of acetylcholinesterase inhibitors

The reversible inhibition of acetylcholinesterase (AChE) has become a promising target for the treatment of Alzheimer's disease (AD) which is mainly associated with low in vivo levels of acetylcholine (ACh). The availability of AChE crystal structures with and without a ligand triggered the effort to find a structure-based design of acetylcholinesterase inhibitors (AChEIs) for AD. The major problem observed with the structure-based design was the feeble robustness of the scoring functions toward the correlation of docking scores with inhibitory potencies of known ligands. This prompted us to develop new prediction models using the stepwise regression analysis based on consensus of different docking and their scoring methods (GOLD, LigandFit, and GLIDE). In the present investigation, a dataset of 91 molecules belonging to 9 different structural classes of heterocyclic compounds with an activity range of 0.008 to 281,000 nM was considered for docking studies and development of AChE-specific 3D-QSAR models. The model (M1) developed using consensus of docking scores of scoring functions viz. Glide score, Gold score, Chem score, ASP score, PMF score, and DOCK score was found to be the best (R(2) = 0.938, Q(2) = 0.925, R(pred)(2) = 0.919, R(2)m((overall)) = 0.936) compared to other consensus models. Docking studies revealed that the molecules with proper alignment in the active site gorge and the ability to interact with all the crucial amino acid residues, in particular by forming π-π stacking interactions with Trp84 at the catalytic anionic site (CAS) and Trp279 at peripheral anionic site (PAS), showed augmented potencies with consequent improvement in patient cognition and reduced the formation of senile plaques associated with AD. Further, the descriptors that signify the association of the ligands with the receptor as well as ADME properties of the ligands were also analyzed by means of the set of ligands that have been pre-positioned with respect to a receptor after docking analysis and considered as independent variables to generate a linear model (M3 and M4) using a stepwise multiple linear regression method to get additional insight into the physicochemical requirements for effective binding of ligands with AChE as well as for prediction of AChE inhibition. The developed AChE-specific prediction models (M1-M4) satisfactorily reflect the structure-activity relationship of the existing AChEIs and have all the potential to facilitate the process of design and development of new potent AChEIs.

First report on development of quantitative interspecies structure-carcinogenicity relationship models and exploring discriminatory features for rodent carcinogenicity of diverse organic chemicals using OECD guidelines

Different regulatory agencies in food and drug administration and environmental protection worldwide are employing quantitative structure-activity relationship (QSAR) models to fill the data gaps related with properties of chemicals affecting the environment and human health. Carcinogenicity is a toxicity endpoint of major concern in recent times. Interspecies toxicity correlations may provide a tool for estimating sensitivity towards toxic chemical exposure with known levels of uncertainty for a diversity of wildlife species. In this background, we have developed quantitative interspecies structure-carcinogenicity correlation models for rat and mouse [rodent species according to the Organization for Economic Cooperation and Development (OECD) guidelines] based on the carcinogenic potential of 166 organic chemicals with wide diversity of molecular structures, spanning a large number of chemical classes and biological mechanisms. All the developed models have been assessed according to the OECD principles for the validation of QSAR models. Consensus predictions for carcinogenicity of the individual compounds are presented here for any one species when the data for the other species are available. Informative illustrations of the contributing structural fragments of chemicals which are responsible for specific carcinogenicity endpoints are identified by the developed models. The models have also been used to predict mouse carcinogenicities of 247 organic chemicals (for which rat carcinogenicities are present) and rat carcinogenicities of 150 chemicals (for which mouse carcinogenicities are present). Discriminatory features for rat and mouse carcinogenicity values have also been explored.

QSPR-based estimation of the half-lives for polychlorinated biphenyl congeners

In this study, the depuration half-lives of 62 polychlorinated biphenyl (PCB) congeners in juvenile rainbow trout (Oncorhynchus mykiss) were estimated from their structural features based on QSPR methodology. A genetic algorithm (GA) was applied as a variable subset selection strategy and QSPR models established upon multiple linear regression (MLR), multilayer perceptron neural network (MLP NN) and support vector regression (SVR) procedures. Robustness and predictive stability of the constructed models were evaluated through internal and external validation methods. The high numerical values of [Formula: see text] and [Formula: see text], and low RMSE in the case of the MLP NN model, confirm the supremacy of this model as well as nonlinear dependency of molecular structural features to the PCB congeners half-lives. In the best developed QSPR model the following four descriptors are used; lopping centric index (Lop), mean topological charge index of order 1 (JGI1), Geary autocorrelation lag-8/weighted by atomic Sanderson electronegativities (GATS8e) and highest eigenvalue of Burden matrix/weighted by atomic masses (BEHm3). Analysis of the descriptors involved in these models revealed that 2D molecular structural features, compactness and electronegativities are the main factors contributing to the half-lives of PCBs. The structural information presented in this work can be used for further evaluation of half-lives of PCBs and other similar structural compounds in the environment.

Designing hypothesis of substituted benzoxazinones as HIV-1 reverse transcriptase inhibitors: QSAR approach

A linear quantitative structure activity relationship (QSAR) model is presented for predicting human immunodeficiency virus-1 (HIV-1) reverse transcriptase enzyme inhibition. The 2D QSAR and 3D-QSAR models were developed by stepwise multiple linear regression, partial least square (PLS) regression and k-nearest neighbor-molecular field analysis, PLS regression, respectively using a database consisting of 33 recently discovered benzoxazinones. The primary findings of this study is that the number of hydrogen atoms, number of (-NH2) group connected with solitary single bond alters the inhibition of HIV-1 reverse transcriptase. Further, presence of electrostatic, hydrophobic and steric field descriptors significantly affects the ability of benzoxazinone derivatives to inhibit HIV-1 reverse transcriptase. The selected descriptors could serve as a primer for the design of novel and potent antagonists of HIV-1 reverse transcriptase.

Real external predictivity of QSAR models: how to evaluate it? Comparison of different validation criteria and proposal of using the concordance correlation coefficient

The main utility of QSAR models is their ability to predict activities/properties for new chemicals, and this external prediction ability is evaluated by means of various validation criteria. As a measure for such evaluation the OECD guidelines have proposed the predictive squared correlation coefficient Q(2)(F1) (Shi et al.). However, other validation criteria have been proposed by other authors: the Golbraikh-Tropsha method, r(2)(m) (Roy), Q(2)(F2) (Schüürmann et al.), Q(2)(F3) (Consonni et al.). In QSAR studies these measures are usually in accordance, though this is not always the case, thus doubts can arise when contradictory results are obtained. It is likely that none of the aforementioned criteria is the best in every situation, so a comparative study using simulated data sets is proposed here, using threshold values suggested by the proponents or those widely used in QSAR modeling. In addition, a different and simple external validation measure, the concordance correlation coefficient (CCC), is proposed and compared with other criteria. Huge data sets were used to study the general behavior of validation measures, and the concordance correlation coefficient was shown to be the most restrictive. On using simulated data sets of a more realistic size, it was found that CCC was broadly in agreement, about 96% of the time, with other validation measures in accepting models as predictive, and in almost all the examples it was the most precautionary. The proposed concordance correlation coefficient also works well on real data sets, where it seems to be more stable, and helps in making decisions when the validation measures are in conflict. Since it is conceptually simple, and given its stability and restrictiveness, we propose the concordance correlation coefficient as a complementary, or alternative, more prudent measure of a QSAR model to be externally predictive.

Molecular Docking Guided Comparative GFA, G/PLS, SVM and ANN Models of Structurally Diverse Dual Binding Site Acetylcholinesterase Inhibitors

Recently discovered 42 AChE inhibitors binding at the catalytic and peripheral anionic site were identified on the basis of molecular docking approach, and its comparative quantitative structure-activity relationship (QSAR) models were developed. These structurally diverse inhibitors were obtained by our previously reported high-throughput in vitro screening technique using 384-well plate's assay based on colorimetric method of Ellman. QSAR models were developed using (i) genetic function algorithm, (ii) genetic partial least squares, (iii) support vector machine and (iv) artificial neural network techniques. The QSAR model robustness and significance was critically assessed using different cross-validation techniques on test data set. The generated QSAR models using thermodynamic, electrotopological and electronic descriptors showed that nonlinear methods are more robust than linear methods, and provide insight into the structural features of compounds that are important for AChE inhibition.

On some novel extended topochemical atom (ETA) parameters for effective encoding of chemical information and modelling of fundamental physicochemical properties

Extended topochemical atom (ETA) indices developed by our group have been extensively applied in our previous reports for toxicity and ecotoxicity modelling in the field of quantitative structure-activity relationships (QSARs). In the present study these indices have been further explored by defining additional novel parameters to model n-octanol-water partition coefficient (two data sets; n = 168 and 139), water solubility (n = 193), molar refractivity (n = 166), and aromatic substituent constants π, MR, σ (m), and σ (p) (n = 99). All the models developed in the present study have undergone rigorous internal and external validation tests and the models have high statistical significance and prediction potential. In terms of Q² and r² values the models developed for the datasets of whole molecules are better than those previously reported, with topochemically arrived unique (TAU) indices on the same datasets of chemicals. An attempt has also been made to develop models using non-ETA topological and information indices. Interestingly, ETA and non-ETA models have been found to have similar predictive capacity.

Allosteric inhibition of the hepatitis C virus NS5B polymerase: in silico strategies for drug discovery and development

Chronic infection by hepatitis C virus (HCV) often leads to severe liver disease including cirrhosis, hepatocellular carcinoma and liver failure. Despite it being more than 20 years since the identification of HCV, the current standard of care for treating the infection is based on aspecific therapy often associated with severe side effects and low-sustained virological response. Research is ongoing to develop new and better medications, including a broad range of allosteric NS5B polymerase inhibitors. This article reviews traditional computational methodologies and more recently developed in silico strategies aimed at identifying and optimizing non-nucleoside inhibitors targeting allosteric sites of HCV NS5B polymerase. The drug-discovery approaches reviewed could provide take-home lessons for general computer-aided research projects.