Classification of a diverse set of Tetrahymena pyriformis toxicity chemical compounds from molecular descriptors by statistical learning methods

Review of machine learning and deep learning models for toxicity prediction

The ever-increasing number of chemicals has raised public concerns due to their adverse effects on human health and the environment. To protect public health and the environment, it is critical to assess the toxicity of these chemicals. Traditional in vitro and in vivo toxicity assays are complicated, costly, and time-consuming and may face ethical issues. These constraints raise the need for alternative methods for assessing the toxicity of chemicals. Recently, due to the advancement of machine learning algorithms and the increase in computational power, many toxicity prediction models have been developed using various machine learning and deep learning algorithms such as support vector machine, random forest, k-nearest neighbors, ensemble learning, and deep neural network. This review summarizes the machine learning- and deep learning-based toxicity prediction models developed in recent years. Support vector machine and random forest are the most popular machine learning algorithms, and hepatotoxicity, cardiotoxicity, and carcinogenicity are the frequently modeled toxicity endpoints in predictive toxicology. It is known that datasets impact model performance. The quality of datasets used in the development of toxicity prediction models using machine learning and deep learning is vital to the performance of the developed models. The different toxicity assignments for the same chemicals among different datasets of the same type of toxicity have been observed, indicating benchmarking datasets is needed for developing reliable toxicity prediction models using machine learning and deep learning algorithms. This review provides insights into current machine learning models in predictive toxicology, which are expected to promote the development and application of toxicity prediction models in the future.

Chemical rules for optimization of chemical mutagenicity via matched molecular pairs analysis and machine learning methods

Chemical mutagenicity is a serious issue that needs to be addressed in early drug discovery. Over a long period of time, medicinal chemists have manually summarized a series of empirical rules for the optimization of chemical mutagenicity. However, given the rising amount of data, it is getting more difficult for medicinal chemists to identify more comprehensive chemical rules behind the biochemical data. Herein, we integrated a large Ames mutagenicity data set with 8576 compounds to derive mutagenicity transformation rules for reversing Ames mutagenicity via matched molecular pairs analysis. A well-trained consensus model with a reasonable applicability domain was constructed, which showed favorable performance in the external validation set with an accuracy of 0.815. The model was used to assess the generalizability and validity of these mutagenicity transformation rules. The results demonstrated that these rules were of great value and could provide inspiration for the structural modifications of compounds with potential mutagenic effects. We also found that the local chemical environment of the attachment points of rules was critical for successful transformation. To facilitate the use of these mutagenicity transformation rules, we integrated them into ADMETopt2 ( http://lmmd.ecust.edu.cn/admetsar2/admetopt2/ ), a free web server for optimization of chemical ADMET properties. The above-mentioned approach would be extended to the optimization of other toxicity endpoints.

Machine Learning Enabled Structure-Based Drug Repurposing Approach to Identify Potential CYP1B1 Inhibitors

Drug-metabolizing enzyme (DME)-mediated pharmacokinetic resistance of some clinically approved anticancer agents is one of the main reasons for cancer treatment failure. In particular, some commonly used anticancer medicines, including docetaxel, tamoxifen, imatinib, cisplatin, and paclitaxel, are inactivated by CYP1B1. Currently, no approved drugs are available to treat this CYP1B1-mediated inactivation, making the pharmaceutical industries strive to discover new anticancer agents. Because of the extreme complexity and high risk in drug discovery and development, it is worthwhile to come up with a drug repurposing strategy that may solve the resistance problem of existing chemotherapeutics. Therefore, in the current study, a drug repurposing strategy was implemented to find the possible CYP1B1 inhibitors using machine learning (ML) and structure-based virtual screening (SB-VS) approaches. Initially, three different ML models were developed such as support vector machines (SVMs), random forest (RF), and artificial neural network (ANN); subsequently, the best-selected ML model was employed for virtual screening of the selleckchem database to identify potential CYP1B1 inhibitors. The inhibition potency of the obtained hits was judged by analyzing the crucial active site amino acid interactions against CYP1B1. After a thorough assessment of docking scores, binding affinities, as well as binding modes, four compounds were selected and further subjected to in vitro analysis. From the in vitro analysis, it was observed that chlorprothixene, nadifloxacin, and ticagrelor showed promising inhibitory activity toward CYP1B1 in the IC₅₀ range of 0.07-3.00 μM. These new chemical scaffolds can be explored as adjuvant therapies to address CYP1B1-mediated drug-resistance problems.

Multiple machine learning, molecular docking, and ADMET screening approach for identification of selective inhibitors of CYP1B1

Cytochrome P4501B1 is a ubiquitous family protein that is majorly overexpressed in tumors and is responsible for biotransformation-based inactivation of anti-cancer drugs. This inactivation marks the cause of resistance to chemotherapeutics. In the present study, integrated in-silico approaches were utilized to identify selective CYP1B1 inhibitors. To achieve this objective, we initially developed different machine learning models corresponding to two isoforms of the CYP1 family i.e. CYP1A1 and CYP1B1. Subsequently, small molecule databases including ChemBridge, Maybridge, and natural compound library were screened from the selected models of CYP1B1 and CYP1A1. The obtained CYP1B1 inhibitors were further subjected to molecular docking and ADMET analysis. The selectivity of the obtained hits for CYP1B1 over the other isoforms was also judged with molecular docking analysis. Finally, two hits were found to be the most stable which retained key interactions within the active site of CYP1B1 after the molecular dynamics simulations. Novel compound with CYP-D9 and CYP-14 IDs were found to be the most selective CYP1B1 inhibitors which may address the issue of resistance. Moreover, these compounds can be considered as safe agents for further cell-based and animal model studies.Communicated by Ramaswamy H. Sarma.

QSAR modelling study of the bioconcentration factor and toxicity of organic compounds to aquatic organisms using machine learning and ensemble methods

Bioconcentration factors and median lethal concentrations (LC₅₀s) are important when assessing risks posed by organic pollutants to aquatic ecosystems. Various quantitative structure-activity relationship models have been developed to predict bioconcentration factors and classify acute toxicity. In the study, we developed a regression model using Recursive Feature Elimination (RFE) method combined with the Support Vector Machine (SVM) algorithm. We calculated 2D molecular descriptors from a dataset containing 450 diverse chemicals in our regression model. Then we built three ensemble models using three machine learning algorithms and calculated 12 molecular fingerprints from a dataset containing 400 diverse chemicals in our classification models. In the regression model, the R² and R_pred² for the regression model were 0.860 and 0.757, respectively. Other parameters indicated that the regression model made good predictions and could efficiently predict a new set of compounds following standards set by Golbraikh, Tropsha, and Roy. In the classification models, the ensemble-SVM classification model gave an overall accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve of 92.2, 95.1, 86.0, and 0.965, respectively, in a five-fold cross-validation and of 87.3, 92.6, 76.0, and 0.940, respectively, in an external validation. These parameters indicated that our ensemble-SVM model was more stable and gave more accurate predictions than previous models. The model could therefore be used to effectively predict aquatic toxicity and assess risks posed to aquatic ecosystems. We identified several structures most relevant to acute aquatic toxicity through predictions made by the two types of models, and this information may be important to aquatic toxicology experiments and aquatic system management.

Small Molecule Docking of DNA Repair Proteins Associated with Cancer Survival Following PCNA Metagene Adjustment: A Potential Novel Class of Repair Inhibitors

Natural and synthetic small molecules from the NCI Developmental Therapeutics Program (DTP) were employed in molecular dynamics-based docking with DNA repair proteins whose RNA-Seq based expression was associated with overall cancer survival (OS) after adjustment for the PCNA metagene. The compounds employed were required to elicit a sensitive response (vs. resistance) in more than half of the cell lines tested for each cancer. Methodological approaches included peptide sequence alignments and homology modeling for 3D protein structure determination, ligand preparation, docking, toxicity and ADME prediction. Docking was performed for unique lists of DNA repair proteins which predict OS for AML, cancers of the breast, lung, colon, and ovaries, GBM, melanoma, and renal papillary cancer. Results indicate hundreds of drug-like and lead-like ligands with best-pose binding energies less than -6 kcal/mol. Ligand solubility for the top 20 drug-like hits approached lower bounds, while lipophilicity was acceptable. Most ligands were also blood-brain barrier permeable with high intestinal absorption rates. While the majority of ligands lacked positive prediction for HERG channel blockage and Ames carcinogenicity, there was a considerable variation for predicted fathead minnow, honey bee, and Tetrahymena pyriformis toxicity. The computational results suggest the potential for new targets and mechanisms of repair inhibition and can be directly employed for in vitro and in vivo confirmatory laboratory experiments to identify new targets of therapy for cancer survival.

Insights into the Molecular Basis of the Acute Contact Toxicity of Diverse Organic Chemicals in the Honey Bee

Use of chemical pollutants, including pesticides and other industrial chemicals, has resulted in significant risks to the whole ecosystem. Therefore, ecological risk assessment of chemicals is vital and necessary. Since the honey bee (Apis mellifera) is probably among the most exposed species to the polluting chemicals, we focused on the in silico estimation of honey bee toxicity (HBT) of chemicals and the analysis of the relevance of chemical HBT and several key physical-chemical properties and structural characteristics. A total of 40 classification models were developed by combination of five machine learning methods along with seven kinds of fingerprints and a set of molecular descriptors. After 5-fold cross validation and external validation, several models showed good predictive power. The relevance of 12 key physical-chemical properties and chemical HBT was also investigated. Five properties, including AlogP, logD, molecular weight (MW), molecular surface area (MSA), and the number of rotatable bonds (nRTB), indicated positive correlation coefficients with HBT, while molecular solubility (logS) and the number of hydrogen bond donors (nHBD) indicated negative correlation coefficients. Finally, seven privileged substructures responsible for chemical HBT were identified from KRFP and SubFP fingerprints. The results of this study should provide critical information and useful tools for chemical HBT estimation in environmental risk assessment.

Quantitative Regression Models for the Prediction of Chemical Properties by an Efficient Workflow

Rapid safety assessment is more and more needed for the increasing chemicals both in chemical industries and regulators around the world. The traditional experimental methods couldn't meet the current demand any more. With the development of the information technology and the growth of experimental data, in silico modeling has become a practical and rapid alternative for the assessment of chemical properties, especially for the toxicity prediction of organic chemicals. In this study, a quantitative regression workflow was built by KNIME to predict chemical properties. With this regression workflow, quantitative values of chemical properties can be obtained, which is different from the binary-classification model or multi-classification models that can only give qualitative results. To illustrate the usage of the workflow, two predictive models were constructed based on datasets of Tetrahymena pyriformis toxicity and Aqueous solubility. The qcv (2) and qtest (2) of 5-fold cross validation and external validation for both types of models were greater than 0.7, which implies that our models are robust and reliable, and the workflow is very convenient and efficient in prediction of various chemical properties.

Towards cheminformatics-based estimation of drug therapeutic index: Predicting the protective index of anticonvulsants using a new quantitative structure-index relationship approach

The overall efficacy and safety profile of a new drug is partially evaluated by the therapeutic index in clinical studies and by the protective index (PI) in preclinical studies. In-silico predictive methods may facilitate the assessment of these indicators. Although QSAR and QSTR models can be used for predicting PI, their predictive capability has not been evaluated. To test this capability, we developed QSAR and QSTR models for predicting the activity and toxicity of anticonvulsants at accuracy levels above the literature-reported threshold (LT) of good QSAR models as tested by both the internal 5-fold cross validation and external validation method. These models showed significantly compromised PI predictive capability due to the cumulative errors of the QSAR and QSTR models. Therefore, in this investigation a new quantitative structure-index relationship (QSIR) model was devised and it showed improved PI predictive capability that superseded the LT of good QSAR models. The QSAR, QSTR and QSIR models were developed using support vector regression (SVR) method with the parameters optimized by using the greedy search method. The molecular descriptors relevant to the prediction of anticonvulsant activities, toxicities and PIs were analyzed by a recursive feature elimination method. The selected molecular descriptors are primarily associated with the drug-like, pharmacological and toxicological features and those used in the published anticonvulsant QSAR and QSTR models. This study suggested that QSIR is useful for estimating the therapeutic index of drug candidates.

Predicting binding affinities of diverse pharmaceutical chemicals to human serum plasma proteins using QSPR modelling approaches

The prediction of the plasma protein binding (PPB) affinity of chemicals is of paramount significance in the drug development process. In this study, ensemble machine learning-based QSPR models have been established for a four-category classification and PPB affinity prediction of diverse compounds using a large PPB dataset of 930 compounds and in accordance with the OECD guidelines. The structural diversity of the chemicals was tested by the Tanimoto similarity index. The external predictive power of the developed QSPR models was evaluated through internal and external validations. In the QSPR models, XLogP was the most important descriptor. In the test data, the classification QSPR models rendered an accuracy of >93%, while the regression QSPR models yielded r(2) of >0.920 between the measured and predicted PPB affinities, with the root mean squared error <9.77. Values of statistical coefficients derived for the test data were above their threshold limits, thus put a high confidence in this analysis. The QSPR models in this study performed better than any of the previous studies. The results suggest that the developed QSPR models are reliable for predicting the PPB affinity of structurally diverse chemicals. They can be useful for initial screening of candidate molecules in the drug development process.

Synthesis and biological evaluation of a novel class of curcumin analogs as anti-inflammatory agents for prevention and treatment of sepsis in mouse model

A novel class of asymmetric mono-carbonyl analogs of curcumin (AMACs) were synthesized and screened for anti-inflammatory activity. These analogs are chemically stable as characterized by UV absorption spectra. In vitro, compounds 3f, 3m, 4b, and 4d markedly inhibited lipopolysaccharide (LPS)-induced expression of pro-inflammatory cytokines tumor necrosis factor-α and interleukin-6 in a dose-dependent manner, with IC₅₀ values in low micromolar range. In vivo, compound 3f demonstrated potent preventive and therapeutic effects on LPS-induced sepsis in mouse model. Compound 3f downregulated the phosphorylation of extracellular signal-regulated kinase (ERK)1/2 MAPK and suppressed IκBα degradation, which suggests that the possible anti-inflammatory mechanism of compound 3f may be through downregulating nuclear factor kappa binding (NF-κB) and ERK pathways. Also, we solved the crystal structure of compound 3e to confirm the asymmetrical structure. The quantitative structure–activity relationship analysis reveals that the electron-withdrawing substituents on aromatic ring of lead structures could improve activity. These active AMACs represent a new class of anti-inflammatory agents with improved stability, bioavailability, and potency compared to curcumin. Our results suggest that 3f may be further developed as a potential agent for prevention and treatment of sepsis or other inflammation-related diseases.

Qualitative and quantitative structure-activity relationship modelling for predicting blood-brain barrier permeability of structurally diverse chemicals

In this study, structure-activity relationship (SAR) models have been established for qualitative and quantitative prediction of the blood-brain barrier (BBB) permeability of chemicals. The structural diversity of the chemicals and nonlinear structure in the data were tested. The predictive and generalization ability of the developed SAR models were tested through internal and external validation procedures. In complete data, the QSAR models rendered ternary classification accuracy of >98.15%, while the quantitative SAR models yielded correlation (r(2)) of >0.926 between the measured and the predicted BBB permeability values with the mean squared error (MSE) <0.045. The proposed models were also applied to an external new in vitro data and yielded classification accuracy of >82.7% and r(2) > 0.905 (MSE < 0.019). The sensitivity analysis revealed that topological polar surface area (TPSA) has the highest effect in qualitative and quantitative models for predicting the BBB permeability of chemicals. Moreover, these models showed predictive performance superior to those reported earlier in the literature. This demonstrates the appropriateness of the developed SAR models to reliably predict the BBB permeability of new chemicals, which can be used for initial screening of the molecules in the drug development process.

Synthesis and evaluation of a series of novel asymmetrical curcumin analogs for the treatment of inflammation

Curcumin has been reported to possess multiple bioactivities, such as antioxidant, anticancer, and anti-inflammatory properties, however the clinical application of curcumin has been significantly limited by its instability and poor metabolism. Modification of curcumin has led to discovery and development of lots of novel therapeutic candidates. In recent years acute and chronic inflammation has been the focus of numerous studies in various diseases. Here, we synthesized a series of asymmetrical curcumin analogs with high in vitro chemical stability, and their anti-inflammatory activity was evaluated in LPS-stimulated macrophages. According to the bio-screening results and QSAR analysis, these analogs exhibited potent activities against LPS-induced TNF-α and IL-6 release. Among the analogs of the potent anti-inflammatory activity, compounds 3b8 and 3b9 exhibited significant protection and possess enhanced anti-inflammatory activity thereby attenuated the LPS-induced septic death in mice.

Investigating hydrochemistry of groundwater in Indo-Gangetic alluvial plain using multivariate chemometric approaches

Groundwater hydrochemistry of an urban industrial region in Indo-Gangetic plains of north India was investigated. Groundwater samples were collected both from the industrial and non-industrial areas of Kanpur. The hydrochemical data were analyzed using various water quality indices and nonparametric statistical methods. Principal components analysis (PCA) was performed to identify the factors responsible for groundwater contamination. Ensemble learning-based decision treeboost (DTB) models were constructed to develop discriminating and regression functions to differentiate the groundwater hydrochemistry of the three different areas, to identify the responsible factors, and to predict the groundwater quality using selected measured variables. The results indicated non-normal distribution and wide variability of water quality variables in all the study areas, suggesting for nonhomogenous distribution of sources in the region. PCA results showed contaminants of industrial origin dominating in the region. DBT classification model identified pH, redox potential, total-Cr, and λ 254 as the discriminating variables in water quality of the three areas with the average accuracy of 99.51 % in complete data. The regression model predicted the groundwater chemical oxygen demand values exhibiting high correlation with measured values (0.962 in training; 0.918 in test) and the respective low root mean-squared error of 2.24 and 2.01 in training and test arrays. The statistical and chemometric approaches used here suggest that groundwater hydrochemistry differs in the three areas and is dominated by different variables. The proposed methods can be used as effective tools in groundwater management.

Discovery and evaluation of asymmetrical monocarbonyl analogs of curcumin as anti-inflammatory agents

Sepsis is a systemic inflammatory response syndrome and is mainly caused by lipopolysaccharides (LPS) – a component of the cell walls of gram-negative bacteria, via toll-like receptor 4–mitogen-activated protein kinases/nuclear factor-kappa B-dependent proinflammatory signaling pathway. Here, we synthesized 26 asymmetric monocarbonyl analogs of curcumin and evaluated their anti-inflammatory activity by inhibiting the LPS-induced secretion of tumor necrosis factor-α and interleukin-6 in mouse RAW264.7 macrophages. Five active compounds (3a, 3c, 3d, 3j, and 3l) exhibited dose-dependent inhibition against the release of tumor necrosis factor-α and interleukin-6, and they also showed much higher chemical stability than curcumin in vitro. The anti-inflammatory activity of analogs 3a and 3c may be associated with their inhibition of the phosphorylation of extracellular signal-regulated kinase and the activation of nuclear factor-kappa B. In addition, 3c exhibited significant protection against LPS-induced septic death in vivo. These results indicate that asymmetrical monocarbonyl curcumin analogs may be utilized as candidates for the treatment of acute inflammatory diseases.

In silico prediction of toxicity of non-congeneric industrial chemicals using ensemble learning based modeling approaches

Ensemble learning approach based decision treeboost (DTB) and decision tree forest (DTF) models are introduced in order to establish quantitative structure-toxicity relationship (QSTR) for the prediction of toxicity of 1450 diverse chemicals. Eight non-quantum mechanical molecular descriptors were derived. Structural diversity of the chemicals was evaluated using Tanimoto similarity index. Stochastic gradient boosting and bagging algorithms supplemented DTB and DTF models were constructed for classification and function optimization problems using the toxicity end-point in T. pyriformis. Special attention was drawn to prediction ability and robustness of the models, investigated both in external and 10-fold cross validation processes. In complete data, optimal DTB and DTF models rendered accuracies of 98.90%, 98.83% in two-category and 98.14%, 98.14% in four-category toxicity classifications. Both the models further yielded classification accuracies of 100% in external toxicity data of T. pyriformis. The constructed regression models (DTB and DTF) using five descriptors yielded correlation coefficients (R(2)) of 0.945, 0.944 between the measured and predicted toxicities with mean squared errors (MSEs) of 0.059, and 0.064 in complete T. pyriformis data. The T. pyriformis regression models (DTB and DTF) applied to the external toxicity data sets yielded R(2) and MSE values of 0.637, 0.655; 0.534, 0.507 (marine bacteria) and 0.741, 0.691; 0.155, 0.173 (algae). The results suggest for wide applicability of the inter-species models in predicting toxicity of new chemicals for regulatory purposes. These approaches provide useful strategy and robust tools in the screening of ecotoxicological risk or environmental hazard potential of chemicals.

Predicting acute aquatic toxicity of structurally diverse chemicals in fish using artificial intelligence approaches

The research aims to develop global modeling tools capable of categorizing structurally diverse chemicals in various toxicity classes according to the EEC and European Community directives, and to predict their acute toxicity in fathead minnow using set of selected molecular descriptors. Accordingly, artificial intelligence approach based classification and regression models, such as probabilistic neural networks (PNN), generalized regression neural networks (GRNN), multilayer perceptron neural network (MLPN), radial basis function neural network (RBFN), support vector machines (SVM), gene expression programming (GEP), and decision tree (DT) were constructed using the experimental toxicity data. Diversity and non-linearity in the chemicals' data were tested using the Tanimoto similarity index and Brock-Dechert-Scheinkman statistics. Predictive and generalization abilities of various models constructed here were compared using several statistical parameters. PNN and GRNN models performed relatively better than MLPN, RBFN, SVM, GEP, and DT. Both in two and four category classifications, PNN yielded a considerably high accuracy of classification in training (95.85 percent and 90.07 percent) and validation data (91.30 percent and 86.96 percent), respectively. GRNN rendered a high correlation between the measured and model predicted -log LC50 values both for the training (0.929) and validation (0.910) data and low prediction errors (RMSE) of 0.52 and 0.49 for two sets. Efficiency of the selected PNN and GRNN models in predicting acute toxicity of new chemicals was adequately validated using external datasets of different fish species (fathead minnow, bluegill, trout, and guppy). The PNN and GRNN models showed good predictive and generalization abilities and can be used as tools for predicting toxicities of structurally diverse chemical compounds.

Predicting carcinogenicity of diverse chemicals using probabilistic neural network modeling approaches

Robust global models capable of discriminating positive and non-positive carcinogens; and predicting carcinogenic potency of chemicals in rodents were developed. The dataset of 834 structurally diverse chemicals extracted from Carcinogenic Potency Database (CPDB) was used which contained 466 positive and 368 non-positive carcinogens. Twelve non-quantum mechanical molecular descriptors were derived. Structural diversity of the chemicals and nonlinearity in the data were evaluated using Tanimoto similarity index and Brock-Dechert-Scheinkman statistics. Probabilistic neural network (PNN) and generalized regression neural network (GRNN) models were constructed for classification and function optimization problems using the carcinogenicity end point in rat. Validation of the models was performed using the internal and external procedures employing a wide series of statistical checks. PNN constructed using five descriptors rendered classification accuracy of 92.09% in complete rat data. The PNN model rendered classification accuracies of 91.77%, 80.70% and 92.08% in mouse, hamster and pesticide data, respectively. The GRNN constructed with nine descriptors yielded correlation coefficient of 0.896 between the measured and predicted carcinogenic potency with mean squared error (MSE) of 0.44 in complete rat data. The rat carcinogenicity model (GRNN) applied to the mouse and hamster data yielded correlation coefficient and MSE of 0.758, 0.71 and 0.760, 0.46, respectively. The results suggest for wide applicability of the inter-species models in predicting carcinogenic potency of chemicals. Both the PNN and GRNN (inter-species) models constructed here can be useful tools in predicting the carcinogenicity of new chemicals for regulatory purposes.

Prediction of acute aquatic toxicity in Tetrahymena pyriformis--'Eco-Derek', a knowledge-based system approach

A 'proof-of-concept' version of a software tool for making transparent predictions of acute aquatic toxicity has been developed. It is primarily limited to semi-quantitative predictions in one species, the ciliated protozoan, Tetrahymena pyriformis. A freely available system, 'Eco-Derek', was derived by adapting a well-established, knowledge-based structure-activity and reasoning platform (Derek for Windows, Lhasa Limited). The Derek reasoning code was modified to express potency rather than confidence. Structure-activity relationship (SAR) development utilised a curated version of a published dataset, supplemented with the CADASTER Challenge datasets. Forty-five structural alerts were produced. The dependence on log P was examined for each alert and entered into the system as qualitative reasoning rules specifying the predicted potency as Very Low, Low, Moderate, High or Very High. Evaluation studies showed: (a) moderate accuracy for the training set but low accuracy for an external test set; (b) non-linearity in the toxicity-log P relationship for chemicals without identified structural alerts; (c) insufficient differentiation of substituent effects in some of the reactivity-based structural alerts resulting in too few chemicals predicted with Very High toxicity; and (d) the need for additional structural alerts covering polar narcosis and less common reactive or metabolically activated chemical functionality.

Using Pareto points for model identification in predictive toxicology

: Predictive toxicology is concerned with the development of models that are able to predict the toxicity of chemicals. A reliable prediction of toxic effects of chemicals in living systems is highly desirable in cosmetics, drug design or food protection to speed up the process of chemical compound discovery while reducing the need for lab tests. There is an extensive literature associated with the best practice of model generation and data integration but management and automated identification of relevant models from available collections of models is still an open problem. Currently, the decision on which model should be used for a new chemical compound is left to users. This paper intends to initiate the discussion on automated model identification. We present an algorithm, based on Pareto optimality, which mines model collections and identifies a model that offers a reliable prediction for a new chemical compound. The performance of this new approach is verified for two endpoints: IGC50 and LogP. The results show a great potential for automated model identification methods in predictive toxicology.

QSAR classification of metabolic activation of chemicals into covalently reactive species

Metabolic activation of chemicals into covalently reactive species might lead to toxicological consequences such as tissue necrosis, carcinogenicity, teratogenicity, or immune-mediated toxicities. Early prediction of this undesirable outcome can help in selecting candidates with increased chance of success, thus, reducing attrition at all stages of drug development. The ensemble modelling of mixed features was used for the development of a model to classify the metabolic activation of chemicals into covalently reactive species. The effects of the quality of base classifiers and performance measure for sorting were examined. An ensemble model of 13 naive Bayes classifiers was built from a diverse set of 1,479 compounds. The ensemble model was validated internally with five-fold cross validation and it has achieved sensitivity of 67.4% and specificity of 93.4% when tested on the training set. The final ensemble model was made available for public use.

Recent trends in statistical QSAR modeling of environmental chemical toxicity

Quantitative cheminformatics approaches such as QSAR modeling find growing applications in chemical risk assessment. Traditional methods rely on the use of calculated chemical descriptors of molecules and relatively small training sets. However, in recent years, there is a trend toward the increased use of in vitro biological testing approaches to reduce both the length of experimental studies and the animal use for chemical risk assessment. Furthermore, there is also much greater emphasis on model validation using external datasets to enable the reliable use of computational models as part of regulatory decision making. In this chapter, recent trends emphasizing the need for both careful curation of experimental data prior to model development and rigorous model validation are investigated. Furthermore, recent approaches to chemical toxicity prediction that employ both chemical descriptors and in vitro screening data for developing novel hybrid chemical/biological models are being reviewed. Examples of respective application studies that employ novel workflows for model developments are described and recent important efforts by several academic, nonprofit, and industrial groups to start placing both data and, especially, models in the public domain are discussed.

In silico prediction of Tetrahymena pyriformis toxicity for diverse industrial chemicals with substructure pattern recognition and machine learning methods

There is an increasing need for the rapid safety assessment of chemicals by both industries and regulatory agencies throughout the world. In silico techniques are practical alternatives in the environmental hazard assessment. It is especially true to address the persistence, bioaccumulative and toxicity potentials of organic chemicals. Tetrahymena pyriformis toxicity is often used as a toxic endpoint. In this study, 1571 diverse unique chemicals were collected from the literature and composed of the largest diverse data set for T. pyriformis toxicity. Classification predictive models of T. pyriformis toxicity were developed by substructure pattern recognition and different machine learning methods, including support vector machine (SVM), C4.5 decision tree, k-nearest neighbors and random forest. The results of a 5-fold cross-validation showed that the SVM method performed better than other algorithms. The overall predictive accuracies of the SVM classification model with radial basis functions kernel was 92.2% for the 5-fold cross-validation and 92.6% for the external validation set, respectively. Furthermore, several representative substructure patterns for characterizing T. pyriformis toxicity were also identified via the information gain analysis methods.

In silico prediction and screening of gamma-secretase inhibitors by molecular descriptors and machine learning methods

Gamma-secretase inhibitors have been explored for the prevention and treatment of Alzheimer's disease (AD). Methods for prediction and screening of gamma-secretase inhibitors are highly desired for facilitating the design of novel therapeutic agents against AD, especially when incomplete knowledge about the mechanism and three-dimensional structure of gamma-secretase. We explored two machine learning methods, support vector machine (SVM) and random forest (RF), to develop models for predicting gamma-secretase inhibitors of diverse structures. Quantitative analysis of the receiver operating characteristic (ROC) curve was performed to further examine and optimize the models. Especially, the Youden index (YI) was initially introduced into the ROC curve of RF so as to obtain an optimal threshold of probability for prediction. The developed models were validated by an external testing set with the prediction accuracies of SVM and RF 96.48 and 98.83% for gamma-secretase inhibitors and 98.18 and 99.27% for noninhibitors, respectively. The different feature selection methods were used to extract the physicochemical features most relevant to gamma-secretase inhibition. To the best of our knowledge, the RF model developed in this work is the first model with a broad applicability domain, based on which the virtual screening of gamma-secretase inhibitors against the ZINC database was performed, resulting in 368 potential hit candidates.

Bridging the gap between molecular descriptors and mechanism: cases studies by molecular dynamics simulations

In recent years, both classification models and quantitative structure-activity relationships (QSARs) have been developed to discriminate the acute toxicity of polar narcotics and uncouplers. One of fundamental issues is how to select and interpret the molecular descriptors used in both methods. In this work, we first employed support vector machine on a dataset containing 155 polar narcotics and 19 uncouplers to filter the predictive hydrophobic and hydrogen bonding descriptors. Molecular dynamics simulations were then conducted to investigate the behavior of salicylate and pentachlorophenol molecules in the context of a palmitoyl-oleoyl-phosphatidylcholine lipid bilayer. The results demonstrated that their equilibrium properties in the lipid bilayer were closely associated with hydrophobic and hydrogen bonding descriptors. The preferable occupations of these molecules in the lipid bilayer were discussed in terms of their modes of toxic action. The observations from molecular dynamics simulations facilitated to elucidate the mechanism of polar narcotics and uncouplers.

Prediction of chemical carcinogenicity by machine learning approaches

In this paper we report a successful application of machine learning approaches to the prediction of chemical carcinogenicity. Two different approaches, namely a support vector machine (SVM) and artificial neural network (ANN), were evaluated for predicting chemical carcinogenicity from molecular structure descriptors. A diverse set of 844 compounds, including 600 carcinogenic (CG+) and 244 noncarcinogenic (CG-) molecules, was used to estimate the accuracies of these approaches. The database was divided into two sets: the model construction set and the independent test set. Relevant molecular descriptors were selected by a hybrid feature selection method combining Fischer's score and Monte Carlo simulated annealing from a wide set of molecular descriptors, including physiochemical properties, constitutional, topological, and geometrical descriptors. The first model validation method was based a five-fold cross-validation method, in which the model construction set is split into five subsets. The five-fold cross-validation was used to select descriptors and optimise the model parameters by maximising the averaged overall accuracy. The final SVM model gave an averaged prediction accuracy of 90.7% for CG+ compounds, 81.6% for CG- compounds and 88.1% for the overall accuracy, while the corresponding ANN model provided an averaged prediction accuracy of 86.1% for CG+ compounds, 79.3% for CG- compounds and 84.2% for the overall accuracy. These results indicate that the hybrid feature selection method is very efficient and the selected descriptors are truly relevant to the carcinogenicity of compounds. Another model validation method, i.e. a hold-out method, was used to build the classification model using the selected descriptors and the optimised model parameters, in which the whole model construction set was used to build the classification model and the independent test set was used to test the predictive ability of the model. The SVM model gave a prediction accuracy of 87.6% for CG+ compounds, 79.1% for CG- compounds and 85.0% for the overall accuracy. The ANN model gave a prediction accuracy of 85.6% for CG+ compounds, 79.1% for CG- compounds and 83.6% for the overall accuracy. The results indicate that the built models are potentially useful for facilitating the prediction of chemical carcinogenicity of untested compounds.