Since the concept of New Approach Methodologies (NAMs) was introduced, various methods including in vitro approaches, Threshold of Toxicological Concern (TTC), and computational models have gained favor among toxicological assessment experts. With the increasing capabilities of computational models, key toxicological endpoints such as acute toxicity, skin/eye irritation, sensitization, chronic toxicity, genotoxicity, carcinogenicity, reproductive and developmental toxicity, and toxicokinetics can now be predicted using QSAR models.

Related Links

How to Use the TTC Method for Evaluating Impurities in Cosmetic Ingredients in China

On September 24, 2025, China’s National Institutes for Food and Drug Control (NIFDC) publicly solicited comments on the Technical Guidelines for the Dossier Requirements of New Cosmetic Ingredient Registration and Notification (Draft for Comments) (hereinafter referred to as the “Draft for Comments”). The Draft for Comments also proposes that computational modeling should be considered for the prediction of certain toxicological endpoints.

Model Classification

Common QSAR models can be categorized into rule-based models and statistic-based models. Rule-based models are developed based on expert-derived rules and knowledge, while statistic-based models are built using mathematical algorithms such as regression analysis, partial least squares (PLS), and neural networks.

Currently, widely used QSAR software tools, such as the OECD QSAR Toolbox, T.E.S.T., and VEGA, often incorporate both types of models, allowing users to select the most appropriate model depending on their specific needs.

Model Applicability

• The applicability domain (AD) and limitations of the model must be clearly described to enable proper evaluation of whether a specific prediction falls within the model's reliable scope.
• Goodness-of-fit, robustness, and predictive performance should be considered--the model must undergo statistical validation to ensure its reliability and scientific credibility.
• Mechanistic interpretation should be provided whenever possible, to support the biological plausibility of the prediction and enhance regulatory acceptance.

Free QSAR Models for Toxicological Endpoint Prediction

The following is a summary of commonly used free QSAR models that can be applied to the prediction of various toxicological endpoints.

Comparison of Free QSAR Tools from a Functional Perspective

From a functional standpoint, QSAR Toolbox offers the most comprehensive coverage, with predictive capabilities spanning all major toxicological endpoints. Even when a substance falls outside the applicability domain of a QSAR model, data gaps can still be addressed through read-across and other analog-based approaches.

VEGA, as a powerful and user-friendly modeling platform, allows users to obtain predictive results simply by inputting the SMILES structure of the compound, making it highly accessible for routine use. Toxtree, primarily based on rule-based models, is often used in combination with other QSAR tools to enhance the accuracy and reliability of predictions. In addition, T.E.S.T. and the Danish (Q)SAR models provide valuable supplementary predictions, which can help increase the confidence in the overall toxicological assessment.

Successful Applications of QSAR Predictions

To date, the direct use of QSAR predictions for primary ingredients in regulatory submissions remains relatively limited. However, QSAR models have already played a significant role in the assessment of pesticide and pharmaceutical impurities, effectively reducing the need for extensive testing and lowering associated costs.

As part of a Weight of Evidence (WoE) approach, QSAR predictions can serve as supporting evidence, helping to reduce the uncertainty inherent in individual model outputs. When integrated with other lines of evidence, QSAR results contribute to a more robust and scientifically sound toxicological assessment. With the ongoing advancement of New Approach Methodologies (NAMs), the role of QSAR models is expected to expand further, particularly in the context of non-animal testing strategies.

To ensure the reliability and regulatory acceptance of QSAR-based toxicity predictions, it is essential to:

• Carefully evaluate the applicability domain (AD) of the model;
• Assess the confidence level of the prediction;
• Compare results across multiple data sources or models for consistency.

These practices are critical for enhancing the relevance and reliability of QSAR results in the risk assessment of new cosmetic ingredients.

About CIRS

The CIRS cosmetic team is dedicated to ensuring that cosmetic products meet stringent global regulatory standards. It can provide one-stop services covering the whole life-cycle of a personal care product, which includes cosmetic ingredient development, physical/chemical tests, toxicological tests (in vivo & in vitro), efficacy studies (in vivo & in vitro), ingredient registration, and product registration.

Cosmetic services in China:

• China Cosmetic Registration and Filing;
• China New Cosmetic Ingredient Registration and Filing;
• China Cosmetic Ingredient Quality and Safety Information Code Application (NMPA Code);
• In-vitro Safety Evaluation;
• In-vivo Safety Evaluation;
• In-vitro Efficacy Evaluation;
• In-vivo Efficacy Evaluation;
• China Cosmetic Safety Assessment Report;
• China Toothpaste Filing;
• China Disinfectant Notification; 
• China Cosmetic Formula/Label/Claim Review; and
• Customs Clearance

If you need any assistance or have any questions, please get in touch with us via test@cirs-group.com.