REGULATION & COMPLIANCE
In recent years, growing awareness of the effects of chemicals on the environment and human health has led to a profound transformation in how risk assessment is conducted. Increasing attention to sustainability, environmental protection and animal welfare has stimulated the development and adoption of alternatives to animal testing.
In this context, in silico methods— such as( quantitative) structureactivity relationship(( Q) SAR)— represent a crucial evolution in the regulation and management of chemical substances. These computational approaches use predictive mathematical models to forecast the toxic effects of substances without the need for in vivo experiments, thus reducing expensive and time-consuming animal tests.
The increasing complexity of the global chemical landscape requires more efficient strategies to predict and manage environmental impacts. In particular, a key component of EU REACH is to encourage alternative methods to animal testing.
Article 13 stipulates that the use of vertebrate animals to generate data must be avoided when scientifically valid alternative methods are available that provide sufficient safety information, such as in silico methods, existing studies, in vitro methods or data from structurally similar substances( read-across). QSAR models are one of the most established and promising in silico models.
What are( Q) SAR models?
QSAR models are mathematical tools that establish a quantitative relationship between the chemical structure of a substance and its biological or toxicological properties. The underlying assumption is that structurally similar molecules exhibit similar activity. This enables the behaviour of new substances to be predicated on the basis of known data, using algorithms and software, without the need for experimental testing for each individual case.
There are two types of QSAR models. Rule-based expert systems rely on toxicological knowledge-derived rules. Experts identify structural fragments( alerts) associated with specific toxic effects, such as genotoxicity.
Their main advantage is transparency; predictions are backed by mechanistic reasoning and often supported by literature references. However, these models have limitations; they typically have a narrow applicability domain, may not explain activity differences within a chemical class and usually have lower predictive accuracy than statistical models.
Statistical-based systems are based on numerical data and regression, machine learning or data mining techniques trained on a specific dataset. They can discover complex patterns even without full understanding of mechanisms of action. These models can offer highly accurate predictions but are often difficult to interpret and may lack transparency for end users.
After a combined application of both rule-based and statistical-based QSAR models, the outcome is a prediction of a previous established endpoint( e. g., mutagenicity). These results will then be reviewed, as part of a tiered approach, using expert knowledge to assess their validity in conjunction with relevant solid experimental and literature data.
The REACH regulation officially recognises QSAR models as scientifically valid tools, provided they meet specific criteria set by OECD guidelines: a defined endpoint, a known applicability domain, a suitable and reliable dataset, robust validation, and interpretability— that is, understanding the role of descriptors in determining toxic activity. In silico tools are therefore acceptable for regulatory purposes if applied according to these standards.
Effective implementation of in silico methods in environmental risk assessment requires an integrated and tiered approach. This approach begins with initial computational predictive
JUL / AUG 2025 SPECCHEMONLINE. COM
41