Refining of spectrophotometric direct peptide reactivity assay (Spectro-DPRA) for in vitro skin sensitization: A novel predictive model based on decision tree.

The Spectrophotometric Direct Peptide Reactivity Assay (Spectro-DPRA) is a high-throughput in chemico assay designed to evaluate Key Event 1 (KE1)-covalent binding to skin proteins-in the skin sensitization adverse outcome pathway (AOP). Although prior validation studies have demonstrated its predictive capacity, its classification thresholds differ from those established in the standard Direct Peptide Reactivity Assay (DPRA). In this study, we optimized the decision-making framework of Spectro-DPRA utilizing a recursive partitioning algorithm trained on 54 reference chemicals with established Local Lymph Node Assay (LLNA) EC3 values. Four decision tree models were generated; Model#3 employed a classification criterion similar to that of DPRA, while Model#4 defined a criterion based on the depletion rate of cysteine or lysine residues. Both Model#3 and Model# 4 were selected for further refinement and for establishing cut-off values related to skin sensitization potency. Verification with an independent test set comprising 20 chemicals (including 10 OECD TG 442C reference substances and 10 additional test chemicals) demonstrated that both Model#3 and Model#4 achieved overall accuracies exceeding 90 %, sensitivities above 92 %, and specificities of 86 % in skin sensitization hazards identification. These results support the integration of Spectro-DPRA into tiered testing strategies and endorse its utility as a robust, animal-free alternative for skin sensitization hazard and potency assessment.