Consensus of classification trees for skin sensitisation hazard prediction.

Area of Science:

• Toxicology
• Computational Chemistry
• Dermatology

Background:

• European Union regulations ban animal testing for cosmetic ingredients since March 2013.
• Existing in silico and in vitro methods for skin sensitization lack full regulatory acceptance.
• A validated non-animal testing approach is needed to replace traditional animal assays.

Purpose of the Study:

• To develop a defined approach for predicting skin sensitization hazard.
• To create a predictive model using a fixed procedure and multiple data sources.
• To classify chemicals as sensitizers or non-sensitizers without animal testing.

Main Methods:

• Utilized a dataset of in chemico (Direct Peptide Reactivity Assay - DPRA) and in vitro (KeratinoSens™, human Cell Line Activation Test - h-CLAT) data.
• Integrated predictions from multiple computational toxicology software packages.
• Developed a consensus model using two classification trees based on protein reactivity and structural features.

Main Results:

• Classification trees based on in silico predictions demonstrated superior performance in hazard prediction.
• The defined approach achieved high accuracy (0.93), sensitivity (0.98), and specificity (0.85) for 269 chemicals.
• The model provides a confidence measure for each prediction, enhancing its utility.

Conclusions:

• The developed defined approach effectively predicts skin sensitization hazard.
• This non-animal testing strategy aligns with EU cosmetic regulations and reduces reliance on animal assays.
• The model offers a reliable and validated method for assessing skin sensitization potential.