Skin sensitization: Testing the accuracy of the ‘2 of 3’ approach
American Board of Toxicology-certified toxicologist Donald Ward, from the Enhesa Sustainable Chemistry team, shares new research on skin sensitization profiling
Skin sensitization is an important issue in chemical safety and regulatory compliance; and finding better ways to test if chemicals can cause skin allergies is a key focus in the science community, especially as industries move away from animal testing.
A recent study I worked on, along with Colleen McLoughlin in the Enhesa toxicology team, takes a closer look at the accuracy of the ‘2of 3’ approach using the OECD QSAR Toolbox to profile chemicals across various classes. It follows a previous view of the OECD toolbox profiler’s performance reported in Guideline No. 497.
Our findings were shared at the 2024 ASCCT Annual Meeting and contribute important insights into the strengths and limitations of this method and the conversation around non-animal testing strategies.
Study design and chemical selection
We conducted this study to evaluate the accuracy of different chemical classes for skin sensitization profiling using the ‘2 of 3’ Approach with OECD QSAR Toolbox. The research focused on using profiling developed from in vitro assays to predict skin sensitization.
The study began with 4,891 Chemical Hazard Assessments (CHAs) from the Enhesa database narrowing down to 1,330 CHAs after excluding certain chemical groups that were not suited for profiling with the OECD toolbox.
While the OECD’s Guideline No. 497 evaluated the same profilers, it relied on a smaller, curated dataset – 56 substances for animal data and 62 for human data – without information on chemical classes. In contrast, our study included a broader and more diverse set of substances. We also compared the results by chemical class and assessed overall performance in terms of accuracy, sensitivity, specificity, and balanced accuracy.
In our study, we used three profilers to screen the chemicals:
- Keratinocyte gene expression;
- Protein Binding Potency h-CLA; and
- Direct Peptide Reactivity Assay (DPRA).
Results: Accuracy varies by chemical class
The results showed an overall accuracy of 78%, with high specificity but lower sensitivity, showing a need for improved detection of weak sensitizers. For instance:
- The Acrylate/Methacrylate category found:
Accuracy of 96.8%, Specificity of 75%, Sensitivity of 100% and a Balanced Accuracy of 87.5%
- The Aliphatic Amines category found:
Accuracy of 67.3%, Specificity of 99%, Sensitivity of 7.5% and a Balanced Accuracy of 53.3% - The Anilines (Hindered) category found:
Accuracy of 40%, Specificity of 0%, Sensitivity of 66.7% and a Balanced Accuracy of 33.3%.
These findings demonstrate the robustness of the ‘2 of 3’ approach for some chemical categories but not others.
The ‘2 of 3’ method achieved 96.8% accuracy and 87.5% balanced accuracy in the Acrylate/Methacrylate category.
Sample size matters: Variability versus Robustness
With larger and more variable datasets, it becomes possible to better assess the robustness and generalizability of prediction models for skin sensitization. In this study, the model was evaluated using a dataset of n = 1330, which is substantially larger than those used in previous studies (n = 19 and n = 244). Larger datasets tend to include a broader range of chemical structures and biological responses, which can challenge model performance but also provide a more realistic picture of how the model might perform in practical applications.
Using the ‘2 of the 3’ consensus method, the model achieved:
- Accuracy: 78%
- Sensitivity: 27%
- Specificity: 98%
- Balanced Accuracy: 63%
In contrast, previous studies with smaller datasets reported:
- Accuracy: 79–89%
- Sensitivity: 74–89%
- Specificity: 67–88%
- Balanced Accuracy: 77–88%
The notably lower sensitivity in the larger dataset indicates that the model was much better at identifying non-sensitizers than sensitizers, resulting in a higher rate of false negatives. This decline in sensitivity suggests that while the model performs well under controlled or limited conditions (as seen in smaller datasets), its ability to detect true sensitizers may be compromised when faced with greater chemical diversity and complexity. Therefore, the larger dataset not only provides a more stringent test of model performance but also highlights areas where the model may need refinement.
Larger test sample sizes help identify chemical categories where predictive models perform well and highlight those that require additional data to improve overall predictive accuracy – an essential step for building regulatory confidence.
Implications and next steps
The study highlights the importance of refining profiling methods to enhance the accuracy of skin sensitization predictions, especially for chemicals that haven’t been evaluated in vitro or in vivo. By identifying both the strengths and limitations of the ‘2 of 3’ approach, the research contributes to safer chemical use and supports regulatory compliance.
Future work will focus on addressing the limitations found in this study to improve the sensitivity and overall performance of the profiling approach. This includes exploring other chemical categories used to train the profilers and integrating new data to better detect weak sensitizers and reduce false negatives.
This research emphasizes the need for continuous improvement of profilers using in vitro testing results to ensure comprehensive and accurate hazard assessments.
Transitioning from in vivo to in chemico, in vitro, and ultimately to in silico approaches not only enhances the accuracy and efficiency of identifying skin sensitizers, it also reduces animal use in toxicological testing, which is a crucial step towards creating a safer world.
Find out more about Don and his work.
Related resources
Strategic chemical list screening: Compliance and beyond
Effective chemical list screening is becoming more complex due to more sources to manage across jurisdictions, diverse product types, and the push to ban more substances. Join our upcoming webinar as our experts share best practices for overcoming the most common challenges in chemical list screening.
Stay up to date on sustainable chemistry
Subscribe to our Enhesa Horizon Sustainable Chemistry newsletter for the latest in chemicals and toxicology developments, including regulatory updates and insights from our toxicology and sustainable chemistry experts.