Non-invasive epidermal proteomics and machine learning permits molecular subclassification of psoriasis and eczematous dermatitis

Current approaches to selecting molecularly targeted therapies (biologics and oral small molecules) for immune-mediated skin diseases largely overlook interindividual immunologic heterogeneity, in part due to challenges of sample collection and the lack of broadly accepted biomarkers of therapeutic response. We sought to develop a rapid, minimally invasive method for obtaining and measuring biomarkers from skin and to generate predictive models of therapy response in two common inflammatory skin diseases, psoriasis and eczema. Here we present Detergent-based Immune Profiling System (DIPS), which enables painless and non-scarring collection of full thickness epidermal protein from skin and is suitable for downstream 45-plex immune protein biomarker analysis. We first developed machine learning models with the goal of accurately distinguishing between eczema (n=55) and psoriasis (n=74). Subsequently, models that correlate biomarker patterns with treatment response or nonresponse to commonly used biologic therapies were developed. We subsequently developed DIPS-Derm, a web-based platform that provides automated diagnostic and treatment predictions from data generated with DIPS. These results support the promise of artificial intelligence (AI)-driven precision dermatology and highlight the clinical potential of DIPS for personalized medicine in inflammatory skin disease.