Case studies on skin sensitization risk assessment: estimating the PoD using artificial neural network-based models for substances with known and unknown structure

Abstract

Non-animal methods for skin sensitization assessment have been developed and adopted as OECD test guidelines. However, no single new approach methodology (NAM) can fully replace animal-based methods, leading to the development of defined approaches like OECD GL497. This study advances quantitative risk assessment (QRA) for skin sensitization using Artificial Neural Network (ANN) models to predict LLNA EC3 values. As a case study, six substances were evaluated using ANN models based on the Direct Peptide Reactivity Assay (DPRA) and the Amino acid Derivative Reactivity Assay (ADRA). These substances included four with known structures (Metol, Dibenzyl Ether, Safranal, and Lyral) and two with unknown structures (Verbena Oil and Oakmoss Extract). Most predicted EC3 values were within a 10-fold range of observed values, demonstrating model reliability. Incorporating ADRA molar and gravimetric method data, ANN models successfully predicted EC3 values for both substances with known and unknown structure, showing their applicability to natural complex substances like botanical extracts. A new skin sensitization risk assessment flow incorporating ANN models is proposed, contributing to the 3Rs by providing a reliable, non-animal method for determining Points of Departure (PoD) and advancing Next Generation Risk Assessment (NGRA) for cosmetic ingredients.