Evaluation of QSAR models for predicting mutagenicity: outcome of the Second Ames/QSAR international challenge project.

Quantitative structure-activity relationship (QSAR) models are powerful in silico tools for predicting the mutagenicity of unstable compounds, impurities and metabolites that are difficult to examine using the Ames test. Ideally, Ames/QSAR models for regulatory use should demonstrate high sensitivity, low false-negative rate and wide coverage of chemical space. To promote superior model development, the Division of Genetics and Mutagenesis, National Institute of Health Sciences, Japan (DGM/NIHS), conducted the Second Ames/QSAR International Challenge Project (2020-2022) as a successor to the First Project (2014-2017), with 21 teams from 11 countries participating. The DGM/NIHS provided a curated training dataset of approximately 12,000 chemicals and a trial dataset of approximately 1,600 chemicals, and each participating team predicted the Ames mutagenicity of each trial chemical using various Ames/QSAR models. The DGM/NIHS then provided the Ames test results for trial chemicals to assist in model improvement. Although overall model performance on the Second Project was not superior to that on the First, models from the eight teams participating in both projects achieved higher sensitivity than models from teams participating in only the Second Project. Thus, these evaluations have facilitated the development of QSAR models.

Quasi-SMILES: Self-consistent models for toxicity of organic chemicals to tadpoles.

Simplified molecular input-line entry systems (SMILES) are the representation of the molecular structure that can be used to establish quantitative structure-property/activity relationships (QSPRs/QSARs) for various endpoints expressed as mathematical functions of the molecular architecture. Quasi-SMILES is extending the traditional SMILES by means of additional symbols that reflect experimental conditions. Using the quasi-SMILES models of toxicity to tadpoles gives the possibility to build up models by taking into account the time of exposure. Toxic effects of experimental situations expressed via 188 quasi-SMILES (the negative logarithm of molar concentrations which lead to lethal 50% tadpoles effected during 12 h, 24 h, 48 h, 72 h, and 96 h) were modelled with good results (the average determination coefficient for the validation sets is about 0.97). In this way, we developed new models for this amphibian endpoint, which is poorly studied.

Monte Carlo technique to study the adsorption affinity of azo dyes by applying new statistical criteria of the predictive potential.

Azo dyes are broadly used in different industries through their chemical stability and ease of synthesis. However, these dyes are usually identified as critical environmental pollutants. Hence, a mathematical model for the adsorption affinity of azo dyes can be applied for solving tasks of medicine and ecology. Quantitative structure-property relationships for the adsorption affinity of azo dyes to a substrate (DAF, kJ/mol) were established using the Monte Carlo method by generating optimal SMILES-based descriptors. The index of ideality of correlation (IIC) and the correlation intensity index (CII) improved the model's predictive potential, especially when they were used simultaneously. The statistical quality of the best model on the validation set was characterized by n = 18, r2 = 0.9468, and RMSE = 1.26 kJ/mol.

Carcinogenicity prediction using the index of ideality of correlation.

Carcinogenicity testing is necessary to protect human health and comply with regulations, but testing it with the traditionally used two-year rodent studies is time-consuming and expensive. In certain cases, such as for impurities, alternative methods may be convenient. Thus there is an urgent need for alternative approaches for reliable and robust assessments of carcinogenicity. The Monte Carlo technique with CORAL software is a tool to tackle this task for unknown compounds using available experimental data for a representative set of compounds. The models can be constructed with the simplified molecular input line entry system without additional physicochemical descriptors. We describe here a model based on a data set of 1167 substances. Matthew's correlation coefficient values for calibration and validation sets are 0.747 and 0.577, respectively. Double bonds between carbon atoms and double bonds of oxygen atoms are the molecular features that indicate the carcinogenic potential of a compound.

Modeling the migration of chemicals from food contact materials to food: The MERLIN-expo/VERMEER toolbox.

Evaluating the migration of chemicals from food contact materials (FCM) into food is a key step in the safety assessment of such materials. In this paper, a simple mechanistic model describing the migration of chemicals from FCM to food was combined with quantitative property-property relationships (QPPRs) for the prediction of diffusion coefficients and FCM-Food partition coefficients. The aim of the present study was to evaluate the performance of these operational models in the prediction of a chemical's concentration in food in contact with a plastic monolayer FCM. A comparison to experimental migration values reported in literature was conducted. Deterministic simulations showed a good match between predicted and experimental values. The tested models can be used to provide insights in the amount and the type of toxicological data that are needed for the safety evaluation of the FCM substance. Uncertainty in QPPRs used for describing the processes of both diffusion in FCM and partition at the FCM-Food interface was included in the analysis. Combining uncertainty in QPPR predictions, it was shown that the third quartile (75th percentile) derived from probabilistic calculations can be used as a conservative value in the prediction of chemical concentration in food, with reasonable safety factors.

The QSAR-search of effective agents towards coronaviruses applying the Monte Carlo method.

Perhaps there is some similarity between the coronavirus of 2017 and the COVID-19. Consequently, a predictive model for the antiviral activity for the Middle East respiratory syndrome coronavirus (MERS-CoV, 2017) could be useful for designing the strategy and tactics in the struggle with coronaviruses in general and with COVID 19 in particular. Quantitative structure-activity relationships (QSARs) of inhibitory activity to MERS-CoV were developed. The index of ideality of correlation was applied to build up these models for the antiviral activity. The statistical quality of the best model is quite good (r2 = 0.84). A mechanistic interpretation of these models based on the molecular features with strong positive (i.e. promoters for endpoint increase) and strong negative (i.e. promoters for endpoint decrease) influence on the inhibitory activity is suggested. A collection of possible biologically active compounds, constructed using data on the above molecular features which are statistically reliable promoters of increase or decrease of the activity, is presented.

The Monte Carlo method to build up models of the hydrolysis half-lives of organic compounds.

The hydrolysis of organic chemicals such as pesticides, pollutants, or drugs can affect the fate and behaviour of environmental contaminants, so it is of interest to evaluate the stability of substances in water for various purposes. For the registration of organic compounds in Europe, information on hydrolysis must be presented. However, the experimental measurements of all chemicals would require enormous resources, and computational models may become attractive. Applying the CORAL software (http://www.insilico.eu/coral) quantitative structure-property relationships (QSPRs) were built up to model hydrolysis. The 2D-optimal descriptor is calculated with so-called correlation weights for attributes of simplified molecular input-line entry systems (SMILES). The correlation weights are obtained as results of the special Monte Carlo optimization. The nature of (five- and six-member) rings is an important component of this approach. Another important component is the atom pair proportions for nitrogen, oxygen, and sulphur. The statistical quality of the best model is: n = 44, r2 = 0.74 (training set); n = 14, r2 = 0.75 (calibration set); and n = 12, r2 = 0.80 (validation set).

EFSA's OpenFoodTox: An open source toxicological database on chemicals in food and feed and its future developments.

Since its creation in 2002, the European Food Safety Authority (EFSA) has produced risk assessments for over 5000 substances in >2000 Scientific Opinions, Statements and Conclusions through the work of its Scientific Panels, Units and Scientific Committee. OpenFoodTox is an open source toxicological database, available both for download and data visualisation which provides data for all substances evaluated by EFSA including substance characterisation, links to EFSA's outputs, applicable legislations regulations, and a summary of hazard identification and hazard characterisation data for human health, animal health and ecological assessments. The database has been structured using OECD harmonised templates for reporting chemical test summaries (OHTs) to facilitate data sharing with stakeholders with an interest in chemical risk assessment, such as sister agencies, international scientific advisory bodies, and others. This manuscript provides a description of OpenFoodTox including data model, content and tools to download and search the database. Examples of applications of OpenFoodTox in chemical risk assessment are discussed including new quantitative structure-activity relationship (QSAR) models, integration into tools (OECD QSAR Toolbox and AMBIT-2.0), assessment of environmental footprints and testing of threshold of toxicological concern (TTC) values for food related compounds. Finally, future developments for OpenFoodTox 2.0 include the integration of new properties, such as physico-chemical properties, exposure data, toxicokinetic information; and the future integration within in silico modelling platforms such as QSAR models and physiologically-based kinetic models. Such structured in vivo, in vitro and in silico hazard data provide different lines of evidence which can be assembled, weighed and integrated using harmonised Weight of Evidence approaches to support the use of New Approach Methodologies (NAMs) in chemical risk assessment and the reduction of animal testing.

Prediction of No Observed Adverse Effect Concentration for inhalation toxicity using Monte Carlo approach.

Ideal correlation is one variable model based on so-called optimal descriptors calculated with simplified molecular input-line entry systems (SMILES). The optimal descriptor is calculated according to the index of ideality of correlation, a new criterion of predictive potential of quantitative structure-property/activity relationships (QSPRs/QSARs). The aim of the present study was the building and estimation of models for inhalation toxicity as No Observed Adverse Effect Concentration (NOAEC) based on the OECD guidelines 413. Three random distributions into the training set and validation set were examined. In practice, a structured training set that contains active training set, passive training set and calibration set is used as the training set. The statistical characteristics of the best model for negative logarithm of NOAEC (pNOAEC) are for training set n = 108, average r 2 = 0.52 + 0.62 + 0.76/3 = 0.63 and for validation set n = 35, r 2 = 0.73.

Modelling quantitative structure activity-activity relationships (QSAARs): auto-pass-pass, a new approach to fill data gaps in environmental risk assessment under the REACH regulation.

Reviewing the toxicological literature for over the past decades, the key elements of QSAR modelling have been the mechanisms of toxic action and chemical classes. As a result, it is often hard to design an acceptable single model for a particular endpoint without clustering compounds. The main aim here was to develop a Pass-Pass Quantitative Structure-Activity-Activity Relationship (PP QSAAR) model for direct prediction of the toxicity of a larger set of compounds, combing the application of an already predicted model for another species, and molecular descriptors. We investigated a large acute toxicity data set of five aquatic organisms, fish, Daphnia magna, and algae from the VEGA-Hub, as well as Tetrahymena pyriformis and Vibrio fischeri. The statistical quality of the models encouraged us to consider this alternative for the prediction of toxicity using interspecies extrapolation QSAAR models without regard to the toxicity mechanism or chemical class. In the case of algae, the use of activity values from a second species did not improve the models. This can be attributed to the weak interspecies relationships, due to different aquatic toxicity mechanisms in species.

QSAR models for biocides: The example of the prediction of Daphnia magna acute toxicity.

Biocides are multi-component products used to control undesired and harmful organisms able to affect human or animal health or to damage natural and manufactured products. Because of their widespread use, aquatic and terrestrial ecosystems could be contaminated by biocides. The environmental impact of biocides is evaluated through eco-toxicological studies with model organisms of terrestrial and aquatic ecosystems. We focused on the development of in silico models for the evaluation of the acute toxicity (EC50) of a set of biocides collected from different sources on the freshwater crustacean Daphnia magna, one of the most widely used model organisms in aquatic toxicology. Toxicological data specific for biocides are limited, so we developed three models for daphnid toxicity using different strategies (linear regression, random forest, Monte Carlo (CORAL)) to overcome this limitation. All models gave satisfactory results in our datasets: the random forest model showed the best results with a determination coefficient r2 = 0.97 and 0.89, respectively, for the training (TS) and the validation sets (VS) while linear regression model and the CORAL model had similar but lower performance (r2 = 0.83 and 0.75, respectively, for TS and VS in the linear regression model and r2 = 0.74 and 0.75 for the CORAL model).

Evaluation of non-commercial models for genotoxicity and carcinogenicity in the assessment of EFSA's databases.

Over the past years, the European Food Safety Authority (EFSA) released to the public domain several databases, with the main objectives of collecting and storing hazard data on the substances considered in EFSA's risk assessment and secondly to serve as a basis for further development of in silico tools such as quantitative structure-activity relationship (QSAR) models. In this work, we evaluated the ability of freely available QSAR models to estimate genotoxicity and carcinogenicity properties and their possible use for screening purposes on three different EFSA's databases. With an accuracy close to 90%, the results showed good capabilities of QSAR models to predict genotoxicity in terms of bacterial reverse mutation test, while statistics for in vivo micronucleus test are not satisfactory (accuracy in the predictions close to 50%). Interestingly, results on the carcinogenicity assessment showed an accuracy in prediction close to 70% for the best models. In addition, an example of the potential application of in silico models is presented in order to provide a preliminary screening of genotoxicity properties of botanicals intended for use as food supplements.

Could deep learning in neural networks improve the QSAR models?

Assessing chemical toxicity is a multidisciplinary process, traditionally involving in vivo, in vitro and in silico tests. Currently, toxicological goal is to reduce new tests on chemicals, exploiting all information yet available. Recent advancements in machine learning and deep neural networks allow computers to automatically mine patterns and learn from data. This technology, applied to (Q)SAR model development, leads to discover by learning the structural-chemical-biological relationships and the emergent properties. Starting from Toxception, a deep neural network predicting activity from the chemical graph image, we designed SmilesNet, a recurrent neural network taking SMILES as the only input. We then integrated the two networks into C-Tox network to make the final classification. Results of our networks, trained on a ~20K molecule dataset with Ames test experimental values, match or even outperform the current state of the art. We also extract knowledge from the networks and compare it with the available mutagenic structural alerts. The advantage over traditional QSAR modelling is that our models automatically extract the features without using descriptors. Nevertheless, the model is successful if large numbers of examples are provided and computation is more complex than in classical methods.

Idealization of correlations between optimal simplified molecular input-line entry system-based descriptors and skin sensitization.

The Index of Ideality of Correlation (IIC) is a new criterion of the predictive potential for quantitative structure-property/activity relationships. The value of the IIC is a mathematical function sensitive to the value of the correlation coefficient and dispersion (expressed via mean absolute error). The IIC has been applied to develop QSAR models for skin sensitization achieving good predictive potential. The 'ideal correlation' is based on elementary fragments of simplified molecular input-line entry system (SMILES) and on the taking into account of the total numbers of nitrogen, oxygen, sulphur and phosphorus in the molecule.

Classification of a Naïve Bayesian Fingerprint model to predict reproductive toxicity$.

Using data from the Leadscope database and Procter and Gamble researchers (1172 compounds after data curation) a new classification model to predict reproductive toxicity was developed. The model is based on Naïve Bayesian methods that use the fingerprint "extended connectivity fingerprint 2". Bits generated by the fingerprint are used from the models as descriptors to discriminate between the two classes. This technique permits the creation of a model without the use of descriptors. After a study on the probability scores, the Naïve Bayesian Fingerprint model shows a good performance on reproductive toxicity. The Matthews Correlation Coefficient value was ≥0.4 in validation. The development of new models to predict complex endpoints such as reproductive toxicity is increasingly requested, with reference also to the REACH legislation in Europe or TSCA in the USA.

A large comparison of integrated SAR/QSAR models of the Ames test for mutagenicity$.

Results from the Ames test are the first outcome considered to assess the possible mutagenicity of substances. Many QSAR models and structural alerts are available to predict this endpoint. From a regulatory point of view, the recommendation from international authorities is to consider the predictions of more than one model and to combine results in order to develop conclusions about the mutagenicity risk posed by chemicals. However, the results of those models are often conflicting, and the existing inconsistency in the predictions requires intelligent strategies to integrate them. In our study, we evaluated different strategies for combining results of models for Ames mutagenicity, starting from a set of 10 diverse individual models, each built on a dataset of around 6000 compounds. The novelty of our study is that we collected a much larger set of about 18,000 compounds and used the new data to build a family of integrated models. These integrations used probabilistic approaches, decision theory, machine learning, and voting strategies in the integration scheme. Results are discussed considering balanced or conservative perspectives, regarding the possible uses for different purposes, including screening of large collection of substances for prioritization.

Predicting acute contact toxicity of pesticides in honeybees (Apis mellifera) through a k-nearest neighbor model.

Ecological risk assessment of plant protection products (PPPs) requires an understanding of both the toxicity and the extent of exposure to assess risks for a range of taxa of ecological importance including target and non-target species. Non-target species such as honey bees (Apis mellifera), solitary bees and bumble bees are of utmost importance because of their vital ecological services as pollinators of wild plants and crops. To improve risk assessment of PPPs in bee species, computational models predicting the acute and chronic toxicity of a range of PPPs and contaminants can play a major role in providing structural and physico-chemical properties for the prioritisation of compounds of concern and future risk assessments. Over the last three decades, scientific advisory bodies and the research community have developed toxicological databases and quantitative structure-activity relationship (QSAR) models that are proving invaluable to predict toxicity using historical data and reduce animal testing. This paper describes the development and validation of a k-Nearest Neighbor (k-NN) model using in-house software for the prediction of acute contact toxicity of pesticides on honey bees. Acute contact toxicity data were collected from different sources for 256 pesticides, which were divided into training and test sets. The k-NN models were validated with good prediction, with an accuracy of 70% for all compounds and of 65% for highly toxic compounds, suggesting that they might reliably predict the toxicity of structurally diverse pesticides and could be used to screen and prioritise new pesticides.

VOC exposures in California early childhood education environments.

Little information exists about exposures to volatile organic compounds (VOCs) in early childhood education (ECE) environments. We measured 38 VOCs in single-day air samples collected in 2010-2011 from 34 ECE facilities serving California children and evaluated potential health risks. We also examined unknown peaks in the GC/MS chromatographs for indoor samples and identified 119 of these compounds using mass spectral libraries. VOCs found in cleaning and personal care products had the highest indoor concentrations (d-limonene and decamethylcyclopentasiloxane [D5] medians: 33.1 and 51.4 µg/m³, respectively). If reflective of long-term averages, child exposures to benzene, chloroform, ethylbenzene, and naphthalene exceeded age-adjusted "safe harbor levels" based on California's Proposition 65 guidelines (10-5 lifetime cancer risk) in 71%, 38%, 56%, and 97% of facilities, respectively. For VOCs without health benchmarks, we used information from toxicological databases and quantitative structure-activity relationship models to assess potential health concerns and identified 12 VOCs that warrant additional evaluation, including a number of terpenes and fragrance compounds. While VOC levels in ECE facilities resemble those in school and home environments, mitigation strategies are warranted to reduce exposures. More research is needed to identify sources and health risks of many VOCs and to support outreach to improve air quality in ECE facilities.

Improving confidence in (Q)SAR predictions under Canada's Chemicals Management Plan - a chemical space approach.

One of the key challenges of Canada's Chemicals Management Plan (CMP) is assessing chemicals with limited/no empirical hazard data for their risk to human health. In some instances, these chemicals have not been tested broadly for their toxicological potency; as such, limited information exists on their potential to induce human health effects following exposure. Although (quantitative) structure activity relationship ((Q)SAR) models are able to generate predictions to address data gaps for certain toxicological endpoints, the confidence in predictions also needs to be addressed. One way to address this issue is to apply a chemical space approach. This approach uses international toxicological databases, for example, those available in the Organisation for Economic Co-operation and Development (OECD) QSAR Toolbox. The approach,assesses a model's ability to predict the potential hazards of chemicals that have limited hazard data that require assessment under the CMP when compared to a larger, data-rich chemical space that is structurally similar to chemicals of interest. This evaluation of a model's predictive ability makes (Q)SAR analysis more transparent and increases confidence in the application of these predictions in a risk-assessment context. Using this approach, predictions for such chemicals obtained from four (Q)SAR models were successfully classified into high, medium and low confidence levels to better inform their use in decision-making.

Results of a round-robin exercise on read-across.

A round-robin exercise was conducted within the CALEIDOS LIFE project. The participants were invited to assess the hazard posed by a substance, applying in silico methods and read-across approaches. The exercise was based on three endpoints: mutagenicity, bioconcentration factor and fish acute toxicity. Nine chemicals were assigned for each endpoint and the participants were invited to complete a specific questionnaire communicating their conclusions. The interesting aspect of this exercise is the justification behind the answers more than the final prediction in itself. Which tools were used? How did the approach selected affect the final answer?