Systematic evaluation of high-throughput PBK modelling strategies for the prediction of intravenous and oral pharmacokinetics in humans.

Physiologically based kinetic (PBK) modelling offers a mechanistic basis for predicting the pharmaco-/toxicokinetics of compounds and thereby provides critical information for integrating toxicity and exposure data to replace animal testing with in vitro or in silico methods. However, traditional PBK modelling depends on animal and human data, which limits its usefulness for non-animal methods. To address this limitation, high-throughput PBK modelling aims to rely exclusively on in vitro and in silico data for model generation. Here, we evaluate a variety of in silico tools and different strategies to parameterise PBK models with input values from various sources in a high-throughput manner. We gather 2000 + publicly available human in vivo concentration-time profiles of 200 + compounds (IV and oral administration), as well as in silico, in vitro and in vivo determined compound-specific parameters required for the PBK modelling of these compounds. Then, we systematically evaluate all possible PBK model parametrisation strategies in PK-Sim and quantify their prediction accuracy against the collected in vivo concentration-time profiles. Our results show that even simple, generic high-throughput PBK modelling can provide accurate predictions of the pharmacokinetics of most compounds (87% of Cmax and 84% of AUC within tenfold). Nevertheless, we also observe major differences in prediction accuracies between the different parameterisation strategies, as well as between different compounds. Finally, we outline a strategy for high-throughput PBK modelling that relies exclusively on freely available tools. Our findings contribute to a more robust understanding of the reliability of high-throughput PBK modelling, which is essential to establish the confidence necessary for its utilisation in Next-Generation Risk Assessment.

Collaborative SAR Modeling and Prospective In Vitro Validation of Oxidative Stress Activation in Human HepG2 Cells.

Oxidative stress is the consequence of an abnormal increase of reactive oxygen species (ROS). ROS are generated mainly during the metabolism in both normal and pathological conditions as well as from exposure to xenobiotics. Xenobiotics can, on the one hand, disrupt molecular machinery involved in redox processes and, on the other hand, reduce the effectiveness of the antioxidant activity. Such dysregulation may lead to oxidative damage when combined with oxidative stress overpassing the cell capacity to detoxify ROS. In this work, a green fluorescent protein (GFP)-tagged nuclear factor erythroid 2-related factor 2 (NRF2)-regulated sulfiredoxin reporter (Srxn1-GFP) was used to measure the antioxidant response of HepG2 cells to a large series of drug and drug-like compounds (2230 compounds). These compounds were then classified as positive or negative depending on cellular response and distributed among different modeling groups to establish structure-activity relationship (SAR) models. A selection of models was used to prospectively predict oxidative stress induced by a new set of compounds subsequently experimentally tested to validate the model predictions. Altogether, this exercise exemplifies the different challenges of developing SAR models of a phenotypic cellular readout, model combination, chemical space selection, and results interpretation.

A KNIME Workflow to Assist the Analogue Identification for Read-Across, Applied to Aromatase Activity.

The reduction and replacement of in vivo tests have become crucial in terms of resources and animal benefits. The read-across approach reduces the number of substances to be tested, exploiting existing experimental data to predict the properties of untested substances. Currently, several tools have been developed to perform read-across, but other approaches, such as computational workflows, can offer a more flexible and less prescriptive approach. In this paper, we are introducing a workflow to support analogue identification for read-across. The implementation of the workflow was performed using a database of azole chemicals with in vitro toxicity data for human aromatase enzymes. The workflow identified analogues based on three similarities: structural similarity (StrS), metabolic similarity (MtS), and mechanistic similarity (McS). Our results showed how multiple similarity metrics can be combined within a read-across assessment. The use of the similarity based on metabolism and toxicological mechanism improved the predictions in particular for sensitivity. Beyond the results predicting a large population of substances, practical examples illustrate the advantages of the proposed approach.

Prediction of the Neurotoxic Potential of Chemicals Based on Modelling of Molecular Initiating Events Upstream of the Adverse Outcome Pathways of (Developmental) Neurotoxicity.

Developmental and adult/ageing neurotoxicity is an area needing alternative methods for chemical risk assessment. The formulation of a strategy to screen large numbers of chemicals is highly relevant due to potential exposure to compounds that may have long-term adverse health consequences on the nervous system, leading to neurodegeneration. Adverse Outcome Pathways (AOPs) provide information on relevant molecular initiating events (MIEs) and key events (KEs) that could inform the development of computational alternatives for these complex effects. We propose a screening method integrating multiple Quantitative Structure-Activity Relationship (QSAR) models. The MIEs of existing AOP networks of developmental and adult/ageing neurotoxicity were modelled to predict neurotoxicity. Random Forests were used to model each MIE. Predictions returned by single models were integrated and evaluated for their capability to predict neurotoxicity. Specifically, MIE predictions were used within various types of classifiers and compared with other reference standards (chemical descriptors and structural fingerprints) to benchmark their predictive capability. Overall, classifiers based on MIE predictions returned predictive performances comparable to those based on chemical descriptors and structural fingerprints. The integrated computational approach described here will be beneficial for large-scale screening and prioritisation of chemicals as a function of their potential to cause long-term neurotoxic effects.

Monte Carlo Models for Sub-Chronic Repeated-Dose Toxicity: Systemic and Organ-Specific Toxicity.

The risk-characterization of chemicals requires the determination of repeated-dose toxicity (RDT). This depends on two main outcomes: the no-observed-adverse-effect level (NOAEL) and the lowest-observed-adverse-effect level (LOAEL). These endpoints are fundamental requirements in several regulatory frameworks, such as the Registration, Evaluation, Authorization and Restriction of Chemicals (REACH) and the European Regulation of 1223/2009 on cosmetics. The RDT results for the safety evaluation of chemicals are undeniably important; however, the in vivo tests are time-consuming and very expensive. The in silico models can provide useful input to investigate sub-chronic RDT. Considering the complexity of these endpoints, involving variable experimental designs, this non-testing approach is challenging and attractive. Here, we built eight in silico models for the NOAEL and LOAEL predictions, focusing on systemic and organ-specific toxicity, looking into the effects on the liver, kidney and brain. Starting with the NOAEL and LOAEL data for oral sub-chronic toxicity in rats, retrieved from public databases, we developed and validated eight quantitative structure-activity relationship (QSAR) models based on the optimal descriptors calculated by the Monte Carlo method, using the CORAL software. The results obtained with these models represent a good achievement, to exploit them in a safety assessment, considering the importance of organ-related toxicity.

Integrated In Silico Models for the Prediction of No-Observed-(Adverse)-Effect Levels and Lowest-Observed-(Adverse)-Effect Levels in Rats for Sub-chronic Repeated-Dose Toxicity.

Repeated-dose toxicity (RDT) is a critical endpoint for hazard characterization of chemicals and is assessed to derive safe levels of exposure for human health. Here we present the first attempt to model simultaneously no-observed-(adverse)-effect level (NO(A)EL) and lowest-observed-(adverse)-effect level (LO(A)EL). Classification and regression models were derived based on rat sub-chronic repeated dose toxicity data for 327 compounds from the Fraunhofer RepDose database. Multi-category classification models were built for both NO(A)EL and LO(A)EL though a consensus of statistics- and fragment-based algorithms, while regression models were based on quantitative relationships between the endpoints and SMILES-based attributes. NO(A)EL and LO(A)EL models were integrated, and predictions were compared to exclude inconsistent values. This strategy improved the performance of single models, leading to R2 greater than 0.70, root-mean-square error (RMSE) lower than 0.60 (for regression models), and accuracy of 0.61-0.73 (for classification models) on the validation set, based on the endpoint and the threshold applied for selecting predictions. This study confirms the effectiveness of the modeling strategy presented here for assessing RDT of chemicals using in silico models.

Zebrafish AC50 modelling: (Q)SAR models to predict developmental toxicity in zebrafish embryo.

Developmental toxicity refers to the occurrence of adverse effects on a developing organism as a consequence of exposure to hazardous chemicals. The assessment of developmental toxicity has become relevant to the safety assessment process of chemicals. The zebrafish embryo developmental toxicology assay is an emerging test used to screen the teratogenic potential of chemicals and it is proposed as a promising test to replace teratogenic assays with animals. Supported by the increased availability of data from this test, the developmental toxicity assay with zebrafish has become an interesting endpoint for the in silico modelling. The purpose of this study was to build up quantitative structure-activity relationship (QSAR) models. In this work, new in silico models for the evaluation of developmental toxicity were built using a well-defined set of data from the ToxCastTM Phase I chemical library on the zebrafish embryo. Categorical and continuous QSAR models were built by gradient boosting machine learning and the Monte Carlo technique respectively, in accordance with Organization for Economic Co-operation and Development principles and their statistical quality was satisfactory. The classification model reached balanced accuracy 0.89 and Matthews correlation coefficient 0.77 on the test set. The regression model reached correlation coefficient R2 0.70 in external validation and leave-one-out cross-validated Q2 0.73 in internal validation.

QSAR Models for Human Carcinogenicity: An Assessment Based on Oral and Inhalation Slope Factors.

Carcinogenicity is a crucial endpoint for the safety assessment of chemicals and products. During the last few decades, the development of quantitative structure-activity relationship ((Q)SAR) models has gained importance for regulatory use, in combination with in vitro testing or expert-based reasoning. Several classification models can now predict both human and rat carcinogenicity, but there are few models to quantitatively assess carcinogenicity in humans. To our knowledge, slope factor (SF), a parameter describing carcinogenicity potential used especially for human risk assessment of contaminated sites, has never been modeled for both inhalation and oral exposures. In this study, we developed classification and regression models for inhalation and oral SFs using data from the Risk Assessment Information System (RAIS) and different machine learning approaches. The models performed well in classification, with accuracies for the external set of 0.76 and 0.74 for oral and inhalation exposure, respectively, and r2 values of 0.57 and 0.65 in the regression models for oral and inhalation SFs in external validation. These models might therefore support regulators in (de)prioritizing substances for regulatory action and in weighing evidence in the context of chemical safety assessments. Moreover, these models are implemented on the VEGA platform and are now freely downloadable online.

QSAR Development for Plasma Protein Binding: Influence of the Ionization State.

PURPOSE: This study explored several strategies to improve the performance of literature QSAR models for plasma protein binding (PPB), such as a suitable endpoint transformation, a correct representation of chemicals, more consistency in the dataset, and a reliable definition of the applicability domain. METHODS: We retrieved human fraction unbound (Fu) data for 670 compounds from the literature and carefully checked them for consistency. Descriptors were calculated taking account of the ionization state of molecules at physiological pH (7.4), in order to better estimate the affinity of molecules to blood proteins. We used different algorithms and chemical descriptors to explore the most suitable strategy for modeling the endpoint. SMILES (simplified molecular input line entry system)-based string descriptors were also tested with the CORAL software (CORelation And Logic). We did an outlier analysis to establish the models to use (or not to use) in case of well recognized families. RESULTS: Internal validation of the selected models returned Q2 values close to 0.60. External validation also gave r2 values always greater than 0.60. The CORAL descriptor based model for √fu was the best, with r2 0.74 in external validation. CONCLUSIONS: Performance in prediction confirmed the robustness of all the derived models and their suitability for real-life purposes, i.e. screening chemicals for their ADMET profiling. Optimization of descriptors can be useful in order to obtain the correct results with a ionized molecule.

QSAR Modeling of ToxCast Assays Relevant to the Molecular Initiating Events of AOPs Leading to Hepatic Steatosis.

Nonalcoholic hepatic steatosis is a worldwide epidemiological concern since it is among the most prominent hepatic diseases. Indeed, research in toxicology and epidemiology has gathered evidence that exposure to endocrine disruptors can perturb cellular homeostasis and cause this disease. Therefore, assessing the likelihood of a chemical to trigger hepatic steatosis is a matter of the utmost importance. However, systematic in vivo testing of all the chemicals humans are exposed to is not feasible for ethical and economical reasons. In this context, predicting the molecular initiating events (MIE) leading to hepatic steatosis by QSAR modeling is an issue of practical relevance in modern toxicology. In this article, we present QSAR models based on random forest classifiers and DRAGON molecular descriptors for the prediction of in vitro assays that are relevant to MIEs leading to hepatic steatosis. These assays were provided by the ToxCast program and proved to be predictive for the detection of chemical-induced steatosis. During the modeling process, special attention was paid to chemical and toxicological data curation. We adopted two modeling strategies (undersampling and balanced random forests) to develop robust QSAR models from unbalanced data sets. The two modeling approaches gave similar results in terms of predictivity, and most of the models satisfy a minimum percentage of correctly predicted chemicals equal to 75%. Finally, and most importantly, the developed models proved to be useful as an effective in silico screening test for hepatic steatosis.

Integrating computational methods to predict mutagenicity of aromatic azo compounds.

Azo dyes have several industrial uses. However, these azo dyes and their degradation products showed mutagenicity, inducing damage in environmental and human systems. Computational methods are proposed as cheap and rapid alternatives to predict the toxicity of azo dyes. A benchmark dataset of Ames data for 354 azo dyes was employed to develop three classification strategies using knowledge-based methods and docking simulations. Results were compared and integrated with three models from the literature, developing a series of consensus strategies. The good results confirm the usefulness of in silico methods as a support for experimental methods to predict the mutagenicity of azo compounds.

Adverse outcome pathways: opportunities, limitations and open questions.

Adverse outcome pathways (AOPs) are a recent toxicological construct that connects, in a formalized, transparent and quality-controlled way, mechanistic information to apical endpoints for regulatory purposes. AOP links a molecular initiating event (MIE) to the adverse outcome (AO) via key events (KE), in a way specified by key event relationships (KER). Although this approach to formalize mechanistic toxicological information only started in 2010, over 200 AOPs have already been established. At this stage, new requirements arise, such as the need for harmonization and re-assessment, for continuous updating, as well as for alerting about pitfalls, misuses and limits of applicability. In this review, the history of the AOP concept and its most prominent strengths are discussed, including the advantages of a formalized approach, the systematic collection of weight of evidence, the linkage of mechanisms to apical end points, the examination of the plausibility of epidemiological data, the identification of critical knowledge gaps and the design of mechanistic test methods. To prepare the ground for a broadened and appropriate use of AOPs, some widespread misconceptions are explained. Moreover, potential weaknesses and shortcomings of the current AOP rule set are addressed (1) to facilitate the discussion on its further evolution and (2) to better define appropriate vs. less suitable application areas. Exemplary toxicological studies are presented to discuss the linearity assumptions of AOP, the management of event modifiers and compensatory mechanisms, and whether a separation of toxicodynamics from toxicokinetics including metabolism is possible in the framework of pathway plasticity. Suggestions on how to compromise between different needs of AOP stakeholders have been added. A clear definition of open questions and limitations is provided to encourage further progress in the field.

Evaluation and comparison of benchmark QSAR models to predict a relevant REACH endpoint: The bioconcentration factor (BCF).

The bioconcentration factor (BCF) is an important bioaccumulation hazard assessment metric in many regulatory contexts. Its assessment is required by the REACH regulation (Registration, Evaluation, Authorization and Restriction of Chemicals) and by CLP (Classification, Labeling and Packaging). We challenged nine well-known and widely used BCF QSAR models against 851 compounds stored in an ad-hoc created database. The goodness of the regression analysis was assessed by considering the determination coefficient (R(2)) and the Root Mean Square Error (RMSE); Cooper's statistics and Matthew's Correlation Coefficient (MCC) were calculated for all the thresholds relevant for regulatory purposes (i.e. 100L/kg for Chemical Safety Assessment; 500L/kg for Classification and Labeling; 2000 and 5000L/kg for Persistent, Bioaccumulative and Toxic (PBT) and very Persistent, very Bioaccumulative (vPvB) assessment) to assess the classification, with particular attention to the models' ability to control the occurrence of false negatives. As a first step, statistical analysis was performed for the predictions of the entire dataset; R(2)>0.70 was obtained using CORAL, T.E.S.T. and EPISuite Arnot-Gobas models. As classifiers, ACD and logP-based equations were the best in terms of sensitivity, ranging from 0.75 to 0.94. External compound predictions were carried out for the models that had their own training sets. CORAL model returned the best performance (R(2)ext=0.59), followed by the EPISuite Meylan model (R(2)ext=0.58). The latter gave also the highest sensitivity on external compounds with values from 0.55 to 0.85, depending on the thresholds. Statistics were also compiled for compounds falling into the models Applicability Domain (AD), giving better performances. In this respect, VEGA CAESAR was the best model in terms of regression (R(2)=0.94) and classification (average sensitivity>0.80). This model also showed the best regression (R(2)=0.85) and sensitivity (average>0.70) for new compounds in the AD but not present in the training set. However, no single optimal model exists and, thus, it would be wise a case-by-case assessment. Yet, integrating the wealth of information from multiple models remains the winner approach.

CORAL: model for no observed adverse effect level (NOAEL).

The in vivo repeated dose toxicity (RDT) test is intended to provide information on the possible risk caused by repeated exposure to a substance over a limited period of time. The measure of the RDT is the no observed adverse effect level (NOAEL) that is the dose at which no effects are observed, i.e., this endpoint indicates the safety level for a substance. The need to replace in vivo tests, as required by some European Regulations (registration, evaluation authorization and restriction of chemicals) is leading to the searching for reliable alternative methods such as quantitative structure-activity relationships (QSAR). Considering the complexity of the RDT endpoint, for which data quality is limited and depends anyway on the study design, the development of QSAR for this endpoint is an attractive task. Starting from a dataset of 140 organic compounds with NOAEL values related to oral short term toxicity in rats, we developed a QSAR model based on optimal descriptors calculated with simplified molecular input-line entry systems and the graph of atomic orbitals by the Monte Carlo method, using CORAL software. Three different splits into the training, calibration, and validation sets are studied. The mechanistic interpretation of these models in terms of molecular fragment with positive or negative contributions to the endpoint is discussed. The probabilistic definition for the domain of applicability is suggested.

Novel clinical phenotypes, drug categorization, and outcome prediction in drug-induced cholestasis: Analysis of a database of 432 patients developed by literature review and machine learning support.

BACKGROUND: Serum transaminases, alkaline phosphatase and bilirubin are common parameters used for DILI diagnosis, classification, and prognosis. However, the relevance of clinical examination, histopathology and drug chemical properties have not been fully investigated. As cholestasis is a frequent and complex DILI manifestation, our goal was to investigate the relevance of clinical features and drug properties to stratify drug-induced cholestasis (DIC) patients, and to develop a prognosis model to identify patients at risk and high-concern drugs. METHODS: DIC-related articles were searched by keywords and Boolean operators in seven databases. Relevant articles were uploaded onto Sysrev, a machine-learning based platform for article review and data extraction. Demographic, clinical, biochemical, and liver histopathological data were collected. Drug properties were obtained from databases or QSAR modelling. Statistical analyses and logistic regressions were performed. RESULTS: Data from 432 DIC patients associated with 52 drugs were collected. Fibrosis strongly associated with fatality, whereas canalicular paucity and ALP associated with chronicity. Drugs causing cholestasis clustered in three major groups. The pure cholestatic pattern divided into two subphenotypes with differences in prognosis, canalicular paucity, fibrosis, ALP and bilirubin. A predictive model of DIC outcome based on non-invasive parameters and drug properties was developed. Results demonstrate that physicochemical (pKa-a) and pharmacokinetic (bioavailability, CYP2C9) attributes impinged on the DIC phenotype and allowed the identification of high-concern drugs. CONCLUSIONS: We identified novel associations among DIC manifestations and disclosed novel DIC subphenotypes with specific clinical and chemical traits. The developed predictive DIC outcome model could facilitate DIC prognosis in clinical practice and drug categorization.

In Silico Models for Repeated-Dose Toxicity (RDT): Prediction of the No Observed Adverse Effect Level (NOAEL) and Lowest Observed Adverse Effect Level (LOAEL) for Drugs.

Many regulatory contexts require the evaluation of repeated-dose toxicity (RDT) studies conducted in laboratory animals. The main outcome of RDT studies is the identification of the no observed adverse effect level (NOAEL) and the lowest observed adverse effect level (LOAEL) that are normally used as point of departure for the establishment of health-based guidance values. Since in vivo RDT studies are expensive and time-consuming, in silico approaches could offer a valuable alternative. However, NOAEL and LOAEL modeling suffer some limitations since they do not refer to a single end point but to several different effects, and the doses used in experimental studies strongly influence the results. Few attempts to model NOAEL and LOAEL have been reported. The available database and models for the prediction of NOAEL and LOAEL are reviewed here.

In Silico Methods for Chromosome Damage.

Due to the link with serious adverse health effects, genotoxicity is an important toxicological endpoint in each regulatory setting with respect to human health, including for pharmaceuticals. To this extent, a compound potential to induce gene mutations as well as chromosome damage needs to be addressed. For chromosome damage, i.e., the induction of structural or numerical chromosome aberrations, several in vitro and in vivo test methods are available. In order to rapidly collect toxicological data without the need for test material, several in silico tools for chromosome damage have been developed over the last years. In this chapter, a battery of freely available in silico chromosome damage prediction tools for chromosome damage is applied on a dataset of pharmaceuticals. Examples of the different outcomes obtained with the in silico battery are provided and briefly discussed. Furthermore, results for coumarin are presented in more detail as a case study. Overall, it can be concluded that although they are in general less developed than those for mutagenicity, in silico tools for chromosome damage can provide valuable information, especially when combined in a battery.

Ligand-based prediction of hERG-mediated cardiotoxicity based on the integration of different machine learning techniques.

Drug-induced cardiotoxicity is a common side effect of drugs in clinical use or under postmarket surveillance and is commonly due to off-target interactions with the cardiac human-ether-a-go-go-related (hERG) potassium channel. Therefore, prioritizing drug candidates based on their hERG blocking potential is a mandatory step in the early preclinical stage of a drug discovery program. Herein, we trained and properly validated 30 ligand-based classifiers of hERG-related cardiotoxicity based on 7,963 curated compounds extracted by the freely accessible repository ChEMBL (version 25). Different machine learning algorithms were tested, namely, random forest, K-nearest neighbors, gradient boosting, extreme gradient boosting, multilayer perceptron, and support vector machine. The application of 1) the best practices for data curation, 2) the feature selection method VSURF, and 3) the synthetic minority oversampling technique (SMOTE) to properly handle the unbalanced data, allowed for the development of highly predictive models (BAMAX = 0.91, AUCMAX = 0.95). Remarkably, the undertaken temporal validation approach not only supported the predictivity of the herein presented classifiers but also suggested their ability to outperform those models commonly used in the literature. From a more methodological point of view, the study put forward a new computational workflow, freely available in the GitHub repository (https://github.com/PDelre93/hERG-QSAR), as valuable for building highly predictive models of hERG-mediated cardiotoxicity.

Integrate mechanistic evidence from new approach methodologies (NAMs) into a read-across assessment to characterise trends in shared mode of action.

Read-across approaches often remain inconclusive as they do not provide sufficient evidence on a common mode of action across the category members. This read-across case study on thirteen, structurally similar, branched aliphatic carboxylic acids investigates the concept of using human-based new approach methods, such as in vitro and in silico models, to demonstrate biological similarity. Five out of the thirteen analogues have preclinical in vivo studies. Three out of them induced lipid accumulation or hypertrophy in preclinical studies with repeated exposure, which leads to the read-across hypothesis that the analogues can potentially induce hepatic steatosis. To confirm the selection of analogues, the expression patterns of the induced differentially expressed genes (DEGs) were analysed in a human liver model. With increasing dose, the expression pattern within the tested analogues got more similar, which serves as a first indication of a common mode of action and suggests differences in the potency of the analogues. Hepatic steatosis is a well-known adverse outcome, for which over 55 adverse outcome pathways have been identified. The resulting adverse outcome pathway (AOP) network, comprised a total 43 MIEs/KEs and enabled the design of an in vitro testing battery. From the AOP network, ten MIEs, early and late KEs were tested to systematically investigate a common mode of action among the grouped compounds. The targeted testing of AOP specific MIE/KEs shows that biological activity in the category decreases with side chain length. A similar trend was evident in measuring liver alterations in zebra fish embryos. However, activation of single MIEs or early KEs at in vivo relevant doses did not necessarily progress to the late KE "lipid accumulation". KEs not related to the read-across hypothesis, testing for example general mitochondrial stress responses in liver cells, showed no trend or biological similarity. Testing scope is a key issue in the design of in vitro test batteries. The Dempster-Shafer decision theory predicted those analogues with in vivo reference data correctly using one human liver model or the CALUX reporter assays. The case study shows that the read-across hypothesis is the key element to designing the testing strategy. In the case of a good mechanistic understanding, an AOP facilitates the selection of reliable human in vitro models to demonstrate a common mode of action. Testing DEGs, MIEs and early KEs served to show biological similarity, whereas the late KEs become important for confirmation, as progression from MIEs to AO is not always guaranteed.

A descriptor-based analysis to highlight the mechanistic rationale of mutagenicity.

Cancer is a main concern for human health and there is a need of alternative methodologies to rapidly screen large quantitative of compounds that may represent a toxicological risk. Here a statistical analyses is performed on a benchmark database of experimental Ames data to identify chemical descriptors discriminating mutagens and non-mutagens. A total of 53 activating and deactivating modulators are identified, that flagged respectively a percentage of mutagen and non-mutagen up to 87%. Modulators are further combined to form synergistic cross-terms, accounting for the effect that combined properties may have on the final toxicity. Exclusion rules are defined as exception to the modulators. Synergistic cross-terms and exclusion rules improve the enrichment of mutagens/non-mutagens with respect of the original abundance in the dataset to values higher than 95%. The external predictivity of modulators and cross-terms reach balanced accuracy up to 0.775 that is analogous to other mutagenicity models from the literature, confirming the suitability of the rules to real-life screening of chemicals. Modulators are discussed for their mechanistic link to mutagenicity. This analysis confirms the key role of some properties (polarizability, shape, mass, presence of reactive functional groups or unsaturated planar systems) as driving elements for the initiation of the mutagenicity.