Development of an Endpoint-Specific Fingerprint to Identify Structurally Similar Chemicals for Skin Sensitization Assessment by Read-Across.

Abstract

Human health risk assessment of cosmetic ingredients must address skin sensitization due to their direct skin application. With the recent shift toward Next Generation Risk Assessment (NGRA), which utilizes animal-free alternative testing methods, the read-across (RAx) approach has gained prominence. RAx relies on information from structural analogues to fill data gaps and determine a point of departure (PoD). A major challenge in RAx is assessing "similarity" between the target chemical and its analogues, typically based on chemical structure, physicochemical properties, metabolic reactivity, in silico prediction profile, and biological activity. Structural similarity, often evaluated using the Tanimoto coefficient (Tc), is a common first step, and it relies on fingerprints that convert chemical structures into machine-readable formats. However, fingerprint choice significantly influences Tc calculations and subsequent analogue selection. Moreover, no guidelines exist for the use of specific fingerprints, and which fingerprint is most suitable for skin sensitization assessment by RAx has not yet been established. This study aimed to develop a novel fingerprint specifically optimized for skin sensitization assessment. A large dataset of chemicals with the murine local lymph node assay (LLNA)-positive data was used for performance evaluation of various fingerprints. Our findings revealed that a novel fingerprint incorporating Protein Binding Alert-based Fingerprint (PBAF) and Klekota-Roth fingerprint (KRFP) features into ToxPrint outperformed others in separating suitable from unsuitable chemical pairs for RAx. This study underscores the importance of endpoint-specific fingerprint development to improve analogue selection for skin sensitization risk assessments.