Study of structure–toxicity relationship by a counterpropagation neural network

Cheminformatics and Machine Learning Approaches to Assess Aquatic Toxicity Profiles of Fullerene Derivatives

Fullerene derivatives (FDs) are widely used in nanomaterials production, the pharmaceutical industry and biomedicine. In the present study, we focused on the potential toxic effects of FDs on the aquatic environment. First, we analyzed the binding affinity of 169 FDs to 10 human proteins (1D6U, 1E3K, 1GOS, 1GS4, 1H82, 1OG5, 1UOM, 2F9Q, 2J0D, 3ERT) obtained from the Protein Data Bank (PDB) and showing high similarity to proteins from aquatic species. Then, the binding activity of 169 FDs to the enzyme acetylcholinesterase (AChE)-as a known target of toxins in fathead minnows and Daphnia magna, causing the inhibition of AChE-was analyzed. Finally, the structural aquatic toxicity alerts obtained from ToxAlert were used to confirm the possible mechanism of action. Machine learning and cheminformatics tools were used to analyze the data. Counter-propagation artificial neural network (CPANN) models were used to determine key binding properties of FDs to proteins associated with aquatic toxicity. Predicting the binding affinity of unknown FDs using quantitative structure-activity relationship (QSAR) models eliminates the need for complex and time-consuming calculations. The results of the study show which structural features of FDs have the greatest impact on aquatic organisms and help prioritize FDs and make manufacturing decisions.

How fullerene derivatives (FDs) act on therapeutically important targets associated with diabetic diseases

•

Five proteins related to diabetic disease were selected from Protein Data Bank.

•

Binding scores were calculated for five proteins with 169 fullerene derivatives.

•

Correlation between drug-like descriptors and binding scores activity was examined.

•

The contribution of descriptors to protein-ligand binding was demonstrated.

•

The QSARs models for prediction of binding scores activity were built.

Fullerene derivatives (FDs) belong to a relatively new family of nano-sized organic compounds. They are widely applied in materials science, pharmaceutical industry, and (bio) medicine. This research focused on the study of FDs in terms of their potential inhibitory effect on therapeutic targets associated with diabetic disease, as well as analysis of protein–ligand binding in order to identify the key binding characteristics of FDs.

Therapeutic drug compounds when entering the biological system usually inevitably encounter and interact with a vast variety of biomolecules that are responsible for many different functions in organisms. Protein biomolecules are the most important functional components and used in this study as target structures. The structures of proteins [(PDB ID: 1BMQ, 1FM6, 1GPB, 1H5U, 1US0)] belonging to the class of anti-diabetes targets were obtained from the Protein Data Bank (PDB). Protein binding activity data (binding scores) were calculated for the dataset of 169 FDs related to these five proteins. Subsequently, the resulting data were analyzed using various machine learning and cheminformatics methods, including artificial neural network algorithms for variable selection and property prediction.

The Quantitative Structure-Activity Relationship (QSAR) models for prediction of binding scores activity were built up according to five Organization for Economic Co-operation and Development (OECD) principles.

All the data obtained can provide important information for further potential use of FDs with different functional groups as promising medical antidiabetic agents. Binding scores activity can be used for ranking of FDs in terms of their inhibitory activity (pharmacological properties) and potential toxicity.

A Comprehensive Cheminformatics Analysis of Structural Features Affecting the Binding Activity of Fullerene Derivatives

Nanostructures like fullerene derivatives (FDs) belong to a new family of nano-sized organic compounds. Fullerenes have found a widespread application in material science, pharmaceutical, biomedical, and medical fields. This fact caused the importance of the study of pharmacological as well as toxicological properties of this relatively new family of chemicals. In this work, a large set of 169 FDs and their binding activity to 1117 proteins was investigated. The structure-based descriptors widely used in drug design (so-called drug-like descriptors) were applied to understand cheminformatics characteristics related to the binding activity of fullerene nanostructures. Investigation of applied descriptors demonstrated that polarizability, topological diameter, and rotatable bonds play the most significant role in the binding activity of FDs. Various cheminformatics methods, including the counter propagation artificial neural network (CPANN) and Kohonen network as visualization tool, were applied. The results of this study can be applied to compose the priority list for testing in risk assessment related to the toxicological properties of FDs. The pharmacologist can filter the data from the heat map to view all possible side effects for selected FDs.

The way to cover prediction for cytotoxicity for all existing nano-sized metal oxides by using neural network method

The regulatory agencies should fulfil the data gap in toxicity for new chemicals including nano-sized compounds, like metal oxides nanoparticles (MeO_x NPs) according to the registration, evaluation, authorisation and restriction of chemicals (REACH) legislation policy. This study demonstrates the perspective capability of neural network models for prediction of cytotoxicity of MeO_x NPs to bacteria Escherichia coli (E. coli) for the widest range of metal oxides extracted from Periodic table. The counter propagation artificial neural network (CP ANN) models for prediction of cytotoxicity of MeO_x NPs for data sets of 17, 36 and 72 metal oxides were employed in the study. The cytotoxicity of studied metal oxide NPs was correlated with (i) χ-metal electronegativity (EN) by Pauling scale and composition of metal oxides characterised by (ii) number of metal atoms in oxide, (iii) number of oxygen atoms in oxide and (iv) charge of metal cation in oxide. The paper describes the models in context of five OECD principles of validation models accepted for regulatory use. The recommendations were done for the minimal number of cytotoxicity tests needs for evaluation of the large set of MeO_x with different oxidation states. The methodology is expected to be useful for potential hazard assessment of MeO_x NPs and prioritisation for further testing and risk assessment.

Structure-activity relationship study of trifluoromethylketone inhibitors of insect juvenile hormone esterase: comparison of several classification methods

Juvenile hormone esterase (JHE) plays a key role in the development and metamorphosis of holometabolous insects. Its inhibitors could possibly be targeted for insect control. Conversely, JHE may also be involved in endocrine disruption by xenobiotics, resulting in detrimental effects in beneficial insects. There is therefore a need to know the structural characteristics of the molecules able to monitor JHE activity, and to develop SAR and QSAR studies to estimate their effectiveness. For a large diverse population of 181 trifluoromethylketones (TFKs) - the most potent JHE inhibitors known to date - we recently proposed a binary classification (active/inactive) using a support vector machine and Codessa structural descriptors. We have now examined, using the same data set and with the same descriptors, the applicability and performance of five other machine learning approaches. These have been shown able to handle high dimensional data (with descriptors possibly irrelevant or redundant) and to cope with complex mechanisms, but without delivering explicit directly exploitable models. Splitting the data into five batches (training set 80%, test set 20%) and carrying out leave-one-out cross-validation, led to good results of comparable performance, consistent with our previous support vector classifier (SVC) results. Accuracy was greater than 0.80 for all approaches. A reduced set of 15 descriptors common to all the investigated approaches showed good predictive ability (confirmed using a three-layer perceptron) and gives some clues regarding a mechanistic interpretation.

Comparison of criteria used to access carcinogenicity in CPANN QSAR models versus the knowledge-based expert system Toxtree

The primary goal of this study was to describe and compare the criteria used to assess carcinogenic activity. The statistically-based predictive quantitative structure-activity relationship (QSAR) models based on the counter propagation artificial neural network (CPANN) algorithm, and knowledge-based expert systems based on a decision tree structural alert (SA) approach (Toxtree application), were considered. The integration of the QSAR (CPANN models) and SAR (Toxtree SA application) approach contributed to the mechanistic understanding of the QSAR model considered. The mapping technique inherent to CPANN Kohonen enables us to relate the similarities or dissimilarities within a congeneric set of chemicals with particular SAs for carcinogenicity. The focus of our investigations was the similarities and dissimilarities of the features used in the QSAR and SAR methods. Due to the complexity of the carcinogenic endpoint, the integration of different approaches allows the models to be improved and provides a valuable technique for evaluating the safety of chemicals.

Integration of QSAR and SAR methods for the mechanistic interpretation of predictive models for carcinogenicity

The knowledge-based Toxtree expert system (SAR approach) was integrated with the statistically based counter propagation artificial neural network (CP ANN) model (QSAR approach) to contribute to a better mechanistic understanding of a carcinogenicity model for non-congeneric chemicals using Dragon descriptors and carcinogenic potency for rats as a response. The transparency of the CP ANN algorithm was demonstrated using intrinsic mapping technique specifically Kohonen maps. Chemical structures were represented by Dragon descriptors that express the structural and electronic features of molecules such as their shape and electronic surrounding related to reactivity of molecules. It was illustrated how the descriptors are correlated with particular structural alerts (SAs) for carcinogenicity with recognized mechanistic link to carcinogenic activity. Moreover, the Kohonen mapping technique enables one to examine the separation of carcinogens and non-carcinogens (for rats) within a family of chemicals with a particular SA for carcinogenicity. The mechanistic interpretation of models is important for the evaluation of safety of chemicals.

COMPARISON OF MULTIPLE LINEAR REGRESSION AND ARTIFICIAL NEURAL NETWORKS IN PREDICTING OCTANOL/WATER PARTITION COEFFICIENT OF A VARIETY OF ORGANIC MOLECULES

For estimating log P values of a group of organic compounds, a back-propagation neural network with a 9–6–1 architecture was developed with optimal learning rate (ε) and momentum (μ) of 0.24 and 0.82, respectively. A collection of 131 organic compounds was chosen as data set that consists of normal hydrocarbons, alcohols, ethers, amines, ketones, acids, benzene derivatives, phenols, and aldehydes. The data set was divided into a training set consisting of 118 molecules and a prediction set consisting of 18 molecules. The most important properties that affect the partition coefficients of organic compounds (surface/volume, dipole moment, and those which are related to electrostatic potentials such as the sum of charges on the carbon atoms) were used as descriptors. These descriptors were obtained using AM1 semiempirical MO method for the gas phase geometries. The descriptors were selected via developing a multiple linear regression analysis. The ANN calculated values of partition coefficients (log Ps) for molecules of the training and prediction sets are in good agreement with the experimental values.

QSAR models for reproductive toxicity and endocrine disruption activity

Reproductive toxicity is an important regulatory endpoint, which is required in registration procedures of chemicals used for different purposes (for example pesticides). The in vivo tests are expensive, time consuming and require large numbers of animals, which must be sacrificed. Therefore an effort is ongoing to develop alternative In vitro and in silico methods to evaluate reproductive toxicity. In this review we describe some modeling approaches. In the first example we describe the CAESAR model for prediction of reproductive toxicity; the second example shows a classification model for endocrine disruption potential based on counter propagation artificial neural networks; the third example shows a modeling of relative binding affinity to rat estrogen receptor, and the fourth one shows a receptor dependent modeling experiment.

Validation of counter propagation neural network models for predictive toxicology according to the OECD principles: a case study

The OECD has proposed five principles for validation of QSAR models used for regulatory purposes. Here we present a case study investigating how these principles can be applied to models based on Kohonen and counter propagation neural networks. The study is based on a counter propagation network model that has been built using toxicity data in fish fathead minnow for 541 compounds. The study demonstrates that most, if not all, of the OECD criteria may be met when modeling using this neural network approach.

Structure-mutagenicity modelling using counter propagation neural networks

The set of 95 aromatic amines and their mutagenic potency was treated with counter propagation neural network, which enables analysis of self-organising maps (SOMs) and also the prediction of mutagenicity. Compounds were described with four classes of descriptors: topostructural (TS), topochemical (TC), geometrical, and quantum chemical (QC). The models were tested on their prediction ability with leave-one-out (LOO) cross-validation method. The squares of correlation coefficient lie between 0.65 and 0.75 and are comparable with models obtained by linear methods. In addition, we analysed self-organising maps and found clusters of structurally similar compounds.

Structure-Mutagenicity Study of 12 Trimethylimidazopyridine Isomers Using Orbital Energies and “Spectrum-like Representation” As Descriptors

The set of 12 trimethylimidazopyridine isomers with mutagenic potency toward two strains of Salmonella was treated in this study. Ten isomers with known mutagenic properties were taken to build the models. Fifteen molecular orbital energies, or a "spectrum-like" representation of 3D structures, were taken as descriptors. As modeling techniques the multiple linear regression and the counter propagation neural network were applied. Models were tested with the recall ability test and the leave-one-out cross-validation tests. For two isomers, which have not been synthesized yet, we report predicted values for both mutagenic potencies obtained with different models. The best models were found when unoccupied molecular orbital energies are among the descriptors.

Partial least squares modelling of the acute toxicity of aliphatic compounds to Tetrahymena pyriformis

The aim of this study was to evaluate a multivariate statistical model, utilising Partial Least Squares (PLS) analysis, for the prediction of the acute toxicity of aliphatic chemicals to the ciliate Tetrahymena pyriformis. A model was developed that was capable of making a prediction regardless the mechanism of toxic action. The toxicity of 476 compounds, possessing different mechanisms of toxic action was considered. A set of 74 descriptors, including the octanol-water partition coefficient, molecular-orbital descriptors, geometrical, topological and connectivity indices, was generated. A three-component, eight-descriptor PLS model was developed. It was validated by a Y-permutation test and by simulation of external prediction for complementary subsets. A comparison with existing class or mechanism-based models, derived on the same data set, was made.

Modeling toxicity by using supervised kohonen neural networks

Counterprogation neural network is shown to be a powerful and suitable tool for the investigation of toxicity. This study mined a data set of 568 chemicals. Two hundred eighty-two objects were used as the training set and 286 as the test set. The final model developed presents high performances on the data set R(2) = 0.83 (R(2) = 0.97 on the training set, R(2) = 0.59 on the test set). This technique distinguishes itself also for the ability to give to the expert two-dimensional maps suitable for the study of the distribution/clustering of the data and the identification of outliers.

The impact of informatics and computational chemistry on synthesis and screening

High-throughput synthesis and screening technologies have enhanced the impact of computational chemistry on the drug discovery process. From the design of targeted, drug-like libraries to 'virtual' optimization of potency, selectivity and ADME/Tox properties, computational chemists are able to efficiently manage costly resources and dramatically shorten drug discovery cycle times. This review will describe some of the successful strategies and applications of state-of-the-art algorithms to enhance drug discovery, as well as key points in the drug discovery process where computational methods can have, and have had, greatest impact.

A study of structure-carcinogenicity relationship for 86 compounds from NTP data base using topological indices as descriptors

86 compounds from NTP carcinogenic potency data base have been used to derive neural network models. Compounds were described with topological indices. Carcinogenicity has been given as a binary quantity--a compound is carcinogenic or non carcinogenic. Several models have been tested with a recognition ability test and with the leave-one-out cross validation method. For the best model the ratio between correct and wrong classifications was 70/30. Furthermore, the model has been used to classify 17 compounds not used for setting of the models. The predicted carcinogenic classes and the neighbors in the neural network influencing the predictions have been discussed.