Non-communicable inflammatory skin diseases comprise clinically meaningful distinct endotypes as identified by non-hypothesized integration of phenotypic and transcriptome data.

BACKGROUND: Non-communicable inflammatory skin diseases (ncISDs) comprise over 100 conditions with overlapping clinical features. Current classification systems are largely based on morphology and do not adequately reflect underlying molecular mechanisms, limiting the implementation of precision therapies. OBJECTIVE: To establish a data-driven, molecular framework for stratifying ncISDs independent of conventional diagnostic categories. METHODS: We integrated 727 skin transcriptomes with clinical metadata from 390 patients across 22 ncISDs and applied unsupervised and machine-learning approaches to identify molecular endotypes and develop a predictive classifier. RESULTS: We identified thirteen distinct molecular endotypes, which segregated into two major groups: seven driven by immune response programs and six by metabolic and epidermal structural pathways. Common diseases such as psoriasis and eczema distributed across multiple endotypes, highlighting substantial molecular heterogeneity. Psoriasis-dominated endotypes showed differential responses to IL-23-, IL-17-, and TNF-targeted therapies, underscoring clinical relevance. A multi-endotype classifier achieved 81.4% accuracy and was validated in independent cohorts. CONCLUSION: These findings define a molecularly informed framework for disease stratification that transcends traditional diagnostic boundaries and enables more precise, mechanism-based therapeutic decision-making across inflammatory diseases.