Introduction

QSAR (Quantitative Structure-Activity Relationship) tools are computational models that predict biological activity and toxicity based on chemical structure, serving as valuable support for human health toxicological risk assessment in several key ways:

  • Screening of one or more chemicals to evaluate physico-chemical properties, environmental fate parameters and toxicological profiles for EU REACH regulatory submissions
  • Grouping of chemicals, data-gap analysis and read-across
  • Prioritisation of chemicals which may require in vivo and/or in vitro testing
  • Defining testing strategies based on in-silico assessment
  • Analysis of potential impurities and degradation products
  • Threshold of toxicological concern (TTC)-based risk assessments

QSAR tools can be used for REACH compliance support, biocides and agrochemicals assessment, persistent, bio-accumulative, toxic (PBT) assessment and, cosmetics regulation (1223/2009) applications.

At CEA, a suite of open source QSAR tools is used. Provided below is a brief description of the four QSAR applications CEA routinely uses. If needed, on a case-by-case basis, proprietary software predictions can be outsourced for completeness in the in-silico assessment of toxicity. Furthermore, CEA is always investigating the expansion of its QSAR suite.

OECD QSAR Toolbox

The freely available OECD QSAR Toolbox (https://www.oecd.org/en/data/tools/oecd-qsar-toolbox.html ), co-owned by OECD and ECHA, supports reproducible and transparent chemical hazard assessment of chemicals towards minimising animal testing and increase mechanistic knowledge on a specific chemical or group of chemicals. It is used for screening, grouping and read-across mostly to support testing strategies. Robust read-across assessments for untested chemicals are encouraged by ECHA and can be accepted as an alternative to toxicity testing. An automated workflow is also available for reliable skin sensitisation prediction based on the direct approach to skin sensitisation (DASS) OECD guideline 497.

The software contains more than 60 databases with over 150,000 chemicals and greater than 3 million experimental data points. This includes a database with experimental results from REACH registrations and an IUCLID plugin.

For chemical profiling and categorisation, the tool uses 73 profiling schemes available for chemical grouping and categorization to facilitate identification of analogue chemicals based on user-selected interaction mechanisms or molecular features. Metabolite assessment can be incorporated at this stage by including any of the five metabolic and degradation simulators. This step enables development of new chemical categories or refinement of existing ones.

As a next step, read-across and trend analysis supports finding structurally and mechanistically defined analogues and chemical categories. The software provides a data matrix for visualisation of the results to compare results appropriately and consistently and allows detailed reports to be created. External QSAR models are incorporated in the tool and can be applied for screening purposes as needed.

VEGA

VEGA (https://www.vegahub.eu/download/) is a feely available software providing more than 90 (Q)SAR models. The tool provides predictive QSAR models to evaluate chemical properties and predict biological activity. VEGA includes models for hydrophobicity (LogP), bio-concentration factor, aquatic toxicity, mutagenicity, carcinogenicity, developmental toxicity and skin sensitisation.

VEGA checks the chemical similarity between the target substance and the substances in the training set of a specific model then makes additional checks specific to the endpoint and the algorithm. For each (Q)SAR model, VEGA employs quantitative measurements to address the applicability domain (AD) which is composed of multiple factors. Predictions on the most similar substances are used to assess whether the prediction is reliable for the target substance. The experimental values for the most similar substances are then compared with the predicted value of the target substance. In this case, the tool compares the agreement between the two values and any potential inconsistencies are indicated to the user. This automated process is intended to help the user specifically address certain points and it allows to filter out predictions with doubts related to the AD. The training sets and QMRF documentation for most models in VEGA are also made public for transparency purposes.

Toxtree

Toxtree (https://toxtree.sourceforge.net/) is an open-source application which is able to estimate toxic hazard by applying a decision tree approach with arbitrary rules. The software serves as a transparent, rule-based model for making toxicity predictions while providing scientifically defensible results for regulatory purposes. The model utilises the Cramer classification scheme as a priority setting tool in the safety assessment of chemicals. The decision tree categorises substances, mainly on the basis of chemical structure and reactivity, into three classes indicating a high (Class III), medium (Class II) or low (Class I) level of concern and corresponding threshold of toxicological concern (TTC). An extended Cramer rule base was introduced in 2009 to extend the framework from 33 to 44 questions to address misclassifications. Toxtree also provides multiple predictions including skin and eye irritation, mutagenicity and carcinogenicity, in vivo micronucleus in rodents, skin sensitisation, DNA and protein binding and cytochrome P450-mediated drug metabolism and metabolites. Its modular design allows users to apply specific decision trees relevant to their assessment needs.

Danish QSAR Database and QSAR Models

The Danish QSAR Database (https://qsarmodels.food.dtu.dk/) represents one of the most comprehensive freely available QSAR prediction platforms, particularly valuable for regulatory screening and early-stage hazard assessment of chemical substances. There is no test information included in the QSAR Database and it only contains predictions rather than experimental data. This screening tool is intended for weight-of-evidence approaches rather than standalone assessments.

The Danish QSAR models (https://qsarmodels.food.dtu.dk/) are a suite of predictive models developed by the Technical University of Denmark (DTU) National Food Institute. The web-based Danish (Q)SAR Models offers users the opportunity to make on-the-fly predictions for user-defined chemical structures by use of more than 30 models developed by DTU in the Leadscope software.

The Danish (Q)SAR Models website uses versions of models such as Leadscope Predictive Data Miner, MultiCASE, MDL QSAR, SciQSAR and CASE Ultra based on the same training sets and modelling approach. The models cover a wide variety of properties including physicochemical, ecotoxicity, environmental fate, ADME (absorption, distribution, metabolism, and excretion), and other toxicological endpoints.

If you need any assistance or have any questions, please get in touch with us via enquiries@cea-res.co.uk .

Enjoyed this article? Avoid missing out on our future news articles by signing up – you can select which topics you are interested in and can unsubscribe at any time.