Development of machine learning models for the prediction of the skin sensitization potential of small organic compounds

Abstract

Allergic contact dermatitis (ACD) is a common and distressful condition among workers and consumers which is induced by the repeated contact of the skin to a skin sensitizing substance [5–7]. To prevent the induction of ACD, a careful risk assessment according the skin sensitization potential or potency of newly developed chemicals and substances is required. Historically, skin sensitization risk assessment was mainly conducted by animal experiments [8]. Currently, it is desired (and partly legally required [9–13]) to assess skin sensitization potential with non-animal alternatives such as in vitro and in chemico assays and computational methods [14, 15]. Compared to testing approaches, computational methods tout several advantages, including reduced testing time, and lower costs. Thus, computational methods are a promising pillar for a non-animal risk assessment of the skin sensitization potential and potency of small molecules. In this thesis, we aim to support the development of reliable and applicable computational tools for the prediction of skin sensitization potential and potency of small molecules. Special emphasis is placed on aspects to increase the models’ usability and acceptance for risk assessment by providing a solid data basis for model development and evaluation, solid measures of reliability and increased interpretability linked to the biological processes of the induction of skin sensitization.