Semi-correlations as a tool to model for skin sensitization.

Area of Science:

• Toxicology and Computational Chemistry
• Biostatistics and Predictive Modeling

Background:

• Traditional methods for assessing skin sensitization involve animal testing, raising ethical concerns and requiring significant resources.
• Developing reliable in silico models for predicting skin sensitization is crucial for reducing animal use and improving safety assessments.
• Semi-correlation offers a statistical method to analyze the relationship between continuous and binary variables, applicable to predictive modeling.

Purpose of the Study:

• To apply semi-correlation analysis for developing binary classification models of skin sensitization.
• To create predictive models for both animal (Local Lymph Node Assay - LLNA) and human (Direct Peptide Reactivity Assay - DPRA and/or Human Cell Line Activation Test - h-CLAT) endpoints.
• To establish a two-level classification strategy: identifying sensitizers versus non-sensitizers, and subsequently categorizing sensitizers as strong or weak.

Main Methods:

• Utilized semi-correlation to establish binary models predicting skin sensitization.
• Developed two-level classification models: Level 1 for 'sensitizer or non-sensitizer' (all compounds), Level 2 for 'strong or weak sensitizer' (sensitizers only).
• Validated model performance using statistical metrics including sensitivity, specificity, accuracy, and Matthew's Correlation Coefficient (MCC).

Main Results:

• Models demonstrated robust statistical performance across different endpoints (LLNA, DPRA/h-CLAT).
• First-level model performance: sensitivity (0.69-0.88), specificity (0.75-0.89), accuracy (0.77-0.87), MCC (0.54-0.57).
• Second-level model performance: sensitivity (0.70-1.0), specificity (0.78-0.83), accuracy (0.77-0.87), MCC (0.54-0.76).

Conclusions:

• Semi-correlation is a viable statistical approach for building predictive models of skin sensitization potency.
• The developed models provide reliable binary classifications for both initial screening and potency assessment.
• This methodology supports the development of alternative testing strategies, reducing reliance on animal models for skin sensitization evaluation.