Quantitative structure-based modeling applied to characterization and prediction of chemical toxicity

Mutagenicity, carcinogenicity, and other end points

Aiming at understanding the structural and physical chemical basis of the biological activity of chemicals, the science of structure-activity relationships has seen dramatic progress in the last decades. Coarse-grain, qualitative approaches (e.g., the structural alerts), and fine-tuned quantitative structure-activity relationship models have been developed and used to predict the toxicological properties of untested chemicals. More recently, a number of approaches and concepts have been developed as support to, and corollary of, the structure-activity methods. These approaches (e.g., chemical relational databases, expert systems, software tools for manipulating the chemical information) have dramatically expanded the reach of the structure-activity work; at present, they are powerful and inescapable tools for computer chemists, toxicologists, and regulators. This chapter, after a general overview of traditional and well-known approaches, gives a detailed presentation of the latter more recent support tools freely available in the public domain.

New public QSAR model for carcinogenicity

BACKGROUND

One of the main goals of the new chemical regulation REACH (Registration, Evaluation and Authorization of Chemicals) is to fulfill the gaps in data concerned with properties of chemicals affecting the human health. (Q)SAR models are accepted as a suitable source of information. The EU funded CAESAR project aimed to develop models for prediction of 5 endpoints for regulatory purposes. Carcinogenicity is one of the endpoints under consideration.

RESULTS

Models for prediction of carcinogenic potency according to specific requirements of Chemical regulation were developed. The dataset of 805 non-congeneric chemicals extracted from Carcinogenic Potency Database (CPDBAS) was used. Counter Propagation Artificial Neural Network (CP ANN) algorithm was implemented. In the article two alternative models for prediction carcinogenicity are described. The first model employed eight MDL descriptors (model A) and the second one twelve Dragon descriptors (model B). CAESAR's models have been assessed according to the OECD principles for the validation of QSAR. For the model validity we used a wide series of statistical checks. Models A and B yielded accuracy of training set (644 compounds) equal to 91% and 89% correspondingly; the accuracy of the test set (161 compounds) was 73% and 69%, while the specificity was 69% and 61%, respectively. Sensitivity in both cases was equal to 75%. The accuracy of the leave 20% out cross validation for the training set of models A and B was equal to 66% and 62% respectively. To verify if the models perform correctly on new compounds the external validation was carried out. The external test set was composed of 738 compounds. We obtained accuracy of external validation equal to 61.4% and 60.0%, sensitivity 64.0% and 61.8% and specificity equal to 58.9% and 58.4% respectively for models A and B.

CONCLUSION

Carcinogenicity is a particularly important endpoint and it is expected that QSAR models will not replace the human experts opinions and conventional methods. However, we believe that combination of several methods will provide useful support to the overall evaluation of carcinogenicity. In present paper models for classification of carcinogenic compounds using MDL and Dragon descriptors were developed. Models could be used to set priorities among chemicals for further testing. The models at the CAESAR site were implemented in java and are publicly accessible.

New public QSAR model for carcinogenicity

BACKGROUND

One of the main goals of the new chemical regulation REACH (Registration, Evaluation and Authorization of Chemicals) is to fulfill the gaps in data concerned with properties of chemicals affecting the human health. (Q)SAR models are accepted as a suitable source of information. The EU funded CAESAR project aimed to develop models for prediction of 5 endpoints for regulatory purposes. Carcinogenicity is one of the endpoints under consideration.

RESULTS

Models for prediction of carcinogenic potency according to specific requirements of Chemical regulation were developed. The dataset of 805 non-congeneric chemicals extracted from Carcinogenic Potency Database (CPDBAS) was used. Counter Propagation Artificial Neural Network (CP ANN) algorithm was implemented. In the article two alternative models for prediction carcinogenicity are described. The first model employed eight MDL descriptors (model A) and the second one twelve Dragon descriptors (model B). CAESAR's models have been assessed according to the OECD principles for the validation of QSAR. For the model validity we used a wide series of statistical checks. Models A and B yielded accuracy of training set (644 compounds) equal to 91% and 89% correspondingly; the accuracy of the test set (161 compounds) was 73% and 69%, while the specificity was 69% and 61%, respectively. Sensitivity in both cases was equal to 75%. The accuracy of the leave 20% out cross validation for the training set of models A and B was equal to 66% and 62% respectively. To verify if the models perform correctly on new compounds the external validation was carried out. The external test set was composed of 738 compounds. We obtained accuracy of external validation equal to 61.4% and 60.0%, sensitivity 64.0% and 61.8% and specificity equal to 58.9% and 58.4% respectively for models A and B.

CONCLUSION

Carcinogenicity is a particularly important endpoint and it is expected that QSAR models will not replace the human experts opinions and conventional methods. However, we believe that combination of several methods will provide useful support to the overall evaluation of carcinogenicity. In present paper models for classification of carcinogenic compounds using MDL and Dragon descriptors were developed. Models could be used to set priorities among chemicals for further testing. The models at the CAESAR site were implemented in java and are publicly accessible.

Counter propagation artificial neural network categorical models for prediction of carcinogenicity for non-congeneric chemicals

One of the main goals of the new chemical regulation REACH (Registration, Evaluation and Authorization of Chemicals) is to fill the gaps on the toxicological properties of chemicals that affect human health. Carcinogenicity is one of the endpoints under consideration. The information obtained from (quantitative) structure-activity relationship ((Q)SAR) models is accepted as an alternative solution to avoid expensive and time-consuming animal tests. The reported results were obtained within the framework of the European project 'Computer Assisted Evaluation of industrial chemical Substances According to Regulations (CAESAR)'. In this article, we demonstrate intermediate results for counter propagation artificial neural network (CP ANN) models for the prediction category of the carcinogenic potency using two-dimensional (2D) descriptors from different software programs. A total of 805 non-congeneric chemicals were extracted from the Carcinogenic Potency Database (CPDBAS). The resulting models had prediction accuracies for internal (training) and external (test) sets as high as 91-93% and 68-70%, respectively. The sensitivity and specificity of the test set were 69-73 and 63-72% correspondingly. High specificity is critical in models for regulatory use that are aimed at ensuring public safety. Thus, the errors that give rise to false negatives are much more relevant. We discuss how we can increase the number of correctly predicted carcinogens using the correlation between the threshold and the values of the sensitivity and specificity.

The application of structure-based assessment to support safety and chemistry diligence to manage genotoxic impurities in active pharmaceutical ingredients during drug development

Starting materials and intermediates used to synthesize pharmaceuticals are reactive in nature and may be present as impurities in the active pharmaceutical ingredient (API) used for preclinical safety studies and clinical trials. Furthermore, starting materials and intermediates may be known or suspected mutagens and/or carcinogens. Therefore, during drug development due diligence need be applied from two perspectives (1) to understand potential mutagenic and carcinogenic risks associated with compounds used for synthesis and (2) to understand the capability of synthetic processes to control genotoxic impurities in the API. Recently, a task force comprised of experts from pharmaceutical industry proposed guidance, with recommendations for classification, testing, qualification and assessing risk of genotoxic impurities. In our experience the proposed structure-based classification, has differentiated 75% of starting materials and intermediates as mutagenic and non-mutagenic with high concordance (92%) when compared with Ames results. Structure-based assessment has been used to identify genotoxic hazards, and prompted evaluation of fate of genotoxic impurities in API. These two assessments (safety and chemistry) culminate in identification of genotoxic impurities known or suspected to exceed acceptable levels in API, thereby triggering actions needed to assure appropriate control and measurement methods are in place. Hypothetical case studies are presented demonstrating this multi-disciplinary approach.

A topological substructural approach applied to the computational prediction of rodent carcinogenicity

The carcinogenic activity has been investigated by using a topological substructural molecular design approach (TOPS-MODE). A discriminant model was developed to predict the carcinogenic and noncarcinogenic activity on a data set of 189 compounds. The percentage of correct classification was 76.32%. The predictive power of the model was validated by three test: an external test set (compounds not used in the develop of the model, with a 72.97% of good classification), a leave-group-out cross-validation procedure (4-fold full cross-validation, removing 20% of compounds in each cycle, with a good prediction of 76.31%) and two external prediction sets (the first and second exercises of the National Toxicology Program). This methodology evidenced that the hydrophobicity increase the carcinogenic activity and the dipole moment of the molecule decrease it; suggesting the capacity of the TOPS-MODE descriptors to estimate this property for new drug candidates. Finally, the positive and negative fragment contributions to the carcinogenic activity were identified (structural alerts) and their potentialities in the lead generation process and in the design of 'safer' chemicals were evaluated.

The use of structure–activity relationship analysis in the food contact notification program

Food contact substances (FCS) include polymers, paper and paperboard, and substances used in their manufacture, that do not impart a technical effect on food. Moreover, FCSs are industrial chemicals generally consumed at dietary concentrations (DC) of less than 1mg/kg food (ppm), and more commonly at less than 0.05 ppm (50 ppb), in the daily diet. As such, many industrial chemicals have been analyzed for toxicological concern, some of which may share structural similarity with FCSs or their constituents, and the majority of these studies are available in the public domain. The DCs of these compounds lend themselves to using structure-activity relationship (SAR) analysis, as the available "expert systems" and use of analogs allows for prediction and management of potential carcinogens. This paper describes the newly implemented food contact notification (FCN) program, the program by which FDA reviews FCSs for safe use, the administrative review of FCSs, the SAR tools available to FDA, and qualitative and quantitative risk assessments using SAR analysis within the regulatory framework of reviewing the safety of FCSs.

Derivation and Validation of Toxicophores for Mutagenicity Prediction

Mutagenicity is one of the numerous adverse properties of a compound that hampers its potential to become a marketable drug. Toxic properties can often be related to chemical structure, more specifically, to particular substructures, which are generally identified as toxicophores. A number of toxicophores have already been identified in the literature. This study aims at increasing the current degree of reliability and accuracy of mutagenicity predictions by identifying novel toxicophores from the application of new criteria for toxicophore rule derivation and validation to a considerably sized mutagenicity dataset. For this purpose, a dataset of 4337 molecular structures with corresponding Ames test data (2401 mutagens and 1936 nonmutagens) was constructed. An initial substructure-search of this dataset showed that most mutagens were detected by applying only eight general toxicophores. From these eight, more specific toxicophores were derived and approved by employing chemical and mechanistic knowledge in combination with statistical criteria. A final set of 29 toxicophores containing new substructures was assembled that could classify the mutagenicity of the investigated dataset with a total classification error of 18%. Furthermore, mutagenicity predictions of an independent validation set of 535 compounds were performed with an error percentage of 15%. Since these error percentages approach the average interlaboratory reproducibility error of Ames tests, which is 15%, it was concluded that these toxicophores can be applied to risk assessment processes and can guide the design of chemical libraries for hit and lead optimization.

Computational prediction of drug toxicity: the case of mutagenicity and carcinogenicity

The removal of pharmaceuticals from the market is a dramatic demonstration of the urgency of implementing better approaches for the early detection of drug toxicity. This paper summarizes the attempts to predict mutagenicity and carcinogenicity from the chemical structure. The limitations of the available commercial systems are pointed out and the urgency of integrating higher levels of chemical knowledge is stressed.:

Assessment of prediction confidence and domain extrapolation of two structure-activity relationship models for predicting estrogen receptor binding activity

Quantitative structure-activity relationship (QSAR) methods have been widely applied in drug discovery, lead optimization, toxicity prediction, and regulatory decisions. Despite major advances in algorithms and software, QSAR models have inherent limitations associated with a size and chemical-structure diversity of the training set, experimental error, and many characteristics of structure representation and correlation algorithms. Whereas excellent fit to the training data may be readily attainable, often models fail to predict accurately chemicals that are outside their domain of applicability. A QSAR's utility and, in the case of regulatory decisions, justification for usage increasingly depend on the ability to quantify a model's potential for predicting unknown chemicals with some known degree of certainty. It is never possible to predict an unknown chemical with absolute certainty. Here we report on two QSAR models based on different data sets for classification of chemicals according to their ability to bind to the estrogen receptor. The models were developed by using a novel QSAR method, Decision Forest, which combines the results of multiple heterogeneous but comparable Decision Tree models to produce a consensus prediction. We used an extensive cross-validation process to define an applicability domain for model predictions based on two quantitative measures: prediction confidence and domain extrapolation. Together, these measures quantify the accuracy of each prediction within and outside of the training domain. Despite being based on large and diverse training sets, both QSAR models had poor accuracy for chemicals within the domain of low confidence, whereas good accuracy was obtained for those within the domain of high confidence. For prediction in the high confidence domain, accuracy was inversely proportional to the degree of domain extrapolation. The model with a larger training set of 1,092, compared with 232 for the other, was more accurate in predicting chemicals at larger domain extrapolation, and could be particularly useful for rapidly prioritizing potential endocrine disruptors from large chemical universe.

Quantitative structure-property relationships for physiologically based pharmacokinetic modeling of volatile organic chemicals in rats

The objective of present study was to develop quantitative structure-property relationships (QSPRs) for the chemical-specific input parameters of rat physiologically based pharmacokinetic (PBPK) models (i.e., blood:air partition coefficient (P(b)), liver:air partition coefficient (P(l)), muscle:air partition coefficient (P(m)), fat:air partition coefficient (P(f)), and hepatic clearance (CL(h))), for simulating the inhalation pharmacokinetics of volatile organic chemicals (VOCs). The literature data on P(b), P(l), P(f), and P(m) for 46 low-molecular-weight VOCs as well as CL(h) for 25 such VOCs primarily metabolized by CYP2E1 (alkanes, haloalkanes, haloethylenes, and aromatic hydrocarbons) were analysed to develop QSPRs. The QSPRs developed in this study were essentially multilinear additive models, which imply that each fragment in the molecular structure has an additive and constant contribution to partition coefficients and hepatic clearance. Most of the values in the calibration set could be reproduced adequately with the QSPR approach, which involved the calculation of the sum of the frequency of occurrence of fragments (CH(3), CH(2), CH, C, C=C, H, Cl, Br, F, benzene ring, and H in benzene ring structure) times the fragment-specific contributions determined in this study. The QSPRs for P(b), P(l), P(m), P(f), and CL(h) were then included within a PBPK model, which only required the specification of the frequency of occurrence of fragments in a molecule along with exposure concentration and duration as input for conducting pharmacokinetic simulations. This QSPR-PBPK model framework facilitated the prediction of the inhalation pharmacokinetics of four VOCs present in the calibration dataset (toluene, dichloromethane, trichloroethylene, and 1,1,1-trichloroethane) and four VOCs that were not part of the calibration set (1,2,4-trimethyl benzene, ethyl benzene, 1,3-dichloropropene, and 2,2-dichloro-1,1,1-trifluoroethane) but that could be described using the molecular fragments investigated in the present study. The QSPRs developed in this study should be potentially useful for providing a first-cut evaluation of the inhalation pharmacokinetics of VOCs prior to experimentation, as long as the number and nature of the fragments do not exceed the ones in the calibration dataset used in this study.

Carcinogenicity of the aromatic amines: from structure-activity relationships to mechanisms of action and risk assessment

Aromatic amines represent one of the most important classes of industrial and environmental chemicals: many of them have been reported to be powerful carcinogens and mutagens, and/or hemotoxicants. Their toxicity has been studied also with quantitative structure-activity relationship (QSAR) methods: these studies are potentially suitable for investigating mechanisms of action and for estimating the toxicity of compounds lacking experimental determinations. In this paper, we first summarized the QSAR models for the rodent carcinogenicity of the aromatic amines. The gradation of potency of the carcinogenic amines depended firstly on their hydrophobicity, and secondly on electronic (reactivity, propensity to be metabolically transformed) and steric properties. On the contrary, the difference between carcinogenic and non-carcinogenic aromatic amines depended mainly on electronic and steric properties. These QSARs can be used directly for estimating the carcinogenicity of aromatic amines. A two-step prediction is possible: (1) estimation of yes/no activity; (2) if the answer from step 1 is yes, then prediction of the degree of potency. The QSARs for rodent carcinogenicity were put in a wider context by comparing them with those for: (a) Salmonella mutagenicity; (b) general toxicity; (c) enzymatic reactions; (d) physical-chemical reactions. This comparative QSAR exercise generated a coherent global picture of the action mechanisms of the aromatic amines. The QSARs for carcinogenicity were similar to those for Salmonella mutagenicity, thus pointing to a similar mechanism of action. On the contrary, the general toxicity QSARs (both in vitro and in vivo systems) were mostly based on hydrophobicity, pointing to an aspecific mechanism of action much simpler than that for carcinogenicity and mutagenicity. The oxidation of the amines (first step in the main metabolic pathway leading to carcinogenic and mutagenic species) had identical QSARs in both enzymatic and physical-chemical systems, thus providing evidence for the link between simple chemical reactions and those in biological systems. The results show that it is possible to generate mechanistically and statistically sound QSAR models for rodent carcinogenicity, and indirectly that the rodent bioassay is a reliable source of good quality data.

Computer systems for the prediction of toxicity: an update

In order to survive in the current economic climate, the pharmaceutical, agrochemical and personal product companies are required to produce large numbers of new, effective products whilst significantly reducing development time and costs. With the advent of combinatorial chemistry and high-throughput screening (HTS), the numbers of new candidate structures coming out of the discovery cycle has increased significantly. This has created a demand for faster screening of the toxicological properties of these candidates. Not surprisingly, computer methods for toxicity prediction offer an attractive solution to this problem because of their ability to screen large numbers of structures even before synthesis has occurred. In this paper the major, commercially available computer software systems for toxicity prediction are discussed together with their main strengths and limitations.

Modelling mutagenicity using properties calculated by computational chemistry

The recent advances in combinatorial chemistry and high throughput screening technologies have led to an explosion in the numbers of possible therapeutic candidates being produced at the early stages of drug discovery. This rapid increase in the number of chemicals to be classified results in a greater need for alternative methods for the prediction of toxicity. Most QSAR models for mutagenicity have been constructed for congeneric series. The prediction requirements of the pharmaceutical industry, however, cover quite diverse chemical structures. This paper reports a study of mutagenicity data for a diverse set of 90 compounds. Good discriminant models have been built for this data set using properties calculated by the techniques of computational chemistry. Jack-knifed (leave one out) predictions for these models are of the order of 85%.

AI and SAR approaches for predicting chemical carcinogenicity: survey and status report

A wide variety of artificial intelligence (AI) and structure-activity relationship (SAR) approaches have been applied to tackling the general problem of predicting rodent chemical carcinogenicity. Given the diversity of chemical structures and mechanisms relative to this endpoint, the shared challenge of these approaches is to accurately delineate classes of active chemicals representing distinct biological and chemical mechanism domains, and within those classes determine the structural features and properties responsible for modulating activity. In the following discussion, we present a survey of AI and SAR approaches that have been applied to the prediction of rodent carcinogenicity, and discuss these in general terms and in the context of the results of two organized prediction exercises (PTE-1 and PTE-2) sponsored by the US National Cancer Institute/National Toxicology Program. Most models participating in these exercises were successful in identifying major structural-alerting classes of active carcinogens, but failed in modeling the more subtle modifiers to activity within those classes. In addition, methods that incorporated mechanism-based reasoning or biological data along with structural information outperformed models limited to structural information exclusively. Finally, a few recent carcinogenicity-modeling efforts are presented illustrating progress in tackling some aspects of the carcinogenicity prediction problem. The first example, a QSAR model for predicting carcinogenic potency of aromatic amines, illustrates that success is possible within well-represented classes of carcinogens. From the second example, a newly developed FDA/OTR MultiCASE model for predicting the carcinogenicity of pharmaceuticals, we conclude that the definitions of biological activity and nature of chemicals in the training set are important determinants of the predictive success and specificity/sensitivity characteristics of a derived model.

Distributed structure-searchable toxicity (DSSTox) public database network: a proposal

The ability to assess the potential genotoxicity, carcinogenicity, or other toxicity of pharmaceutical or industrial chemicals based on chemical structure information is a highly coveted and shared goal of varied academic, commercial, and government regulatory groups. These diverse interests often employ different approaches and have different criteria and use for toxicity assessments, but they share a need for unrestricted access to existing public toxicity data linked with chemical structure information. Currently, there exists no central repository of toxicity information, commercial or public, that adequately meets the data requirements for flexible analogue searching, Structure-Activity Relationship (SAR) model development, or building of chemical relational databases (CRD). The distributed structure-searchable toxicity (DSSTox) public database network is being proposed as a community-supported, web-based effort to address these shared needs of the SAR and toxicology communities. The DSSTox project has the following major elements: (1) to adopt and encourage the use of a common standard file format (structure data file (SDF)) for public toxicity databases that includes chemical structure, text and property information, and that can easily be imported into available CRD applications; (2) to implement a distributed source approach, managed by a DSSTox Central Website, that will enable decentralized, free public access to structure-toxicity data files, and that will effectively link knowledgeable toxicity data sources with potential users of these data from other disciplines (such as chemistry, modeling, and computer science); and (3) to engage public/commercial/academic/industry groups in contributing to and expanding this community-wide, public data sharing and distribution effort. The DSSTox project's overall aims are to effect the closer association of chemical structure information with existing toxicity data, and to promote and facilitate structure-based exploration of these data within a common chemistry-based framework that spans toxicological disciplines.

Commercial toxicology prediction systems: a regulatory perspective

The use of commercial toxicity prediction systems in a regulatory setting must consider both the limitations and capabilities of the methods, as well as the ultimate use of the predictions, e.g. for testing prioritization, screening, or supporting regulatory decisions. Current systems are better suited to hazard identification (i.e. positive identification of activity-conferring features) than to ruling out hazard. Two recent examples (an EPA testing prioritization exercise for water disinfection byproducts and a regulatory action on 2,4,6-tribromophenol) illustrate issues involved in regulatory applications of SAR and commercial prediction systems. The challenge for the future will be to improve technologies for prediction within the constraints of available data, make optimal use of new test data, and better integrate elements of quantitative modeling (QSAR), empirical association, and biological and chemical mechanisms towards the goal of toxicity prediction.