A bioinformatic single-cell and structure-informed framework identifies a baicalin-CA2-keratinocyte state axis in atopic dermatitis.

Abstract

Atopic dermatitis (AD) is characterized by a self-reinforcing loop between epidermal barrier dysfunction and type 2-skewed inflammation; yet the most perturbed keratinocyte states and actionable epidermal targets remain incompletely defined. We integrated pharmacogenomic target mining, complementary machine-learning feature selection (LASSO and SVM-RFE), single-cell state-resolved perturbation analyses (Augur and scDist), and structure-based molecular modeling (molecular docking, MD simulation, and MM-PBSA free energy calculation) to prioritize candidate targets of baicalin in AD. CA2 emerged as a convergent epidermal candidate; scRNA-seq analyses localized CA2-associated transcriptional differences to keratinocytes, with the keratinocyte compartment exhibiting the disease-associated strongest separability and transcriptomic distance, accompanied by enrichment of metabolic reprogramming, epithelial junction and barrier remodeling, and proliferative quiescence gene programs. Structure-based evaluation supported a computationally plausible baicalin-CA2 interaction, with an estimated MM-PBSA binding free energy of -22.082 kcal/mol. Collectively, these findings nominate a computationally supported "baicalin-CA2-Kcs9" axis as a hypothesis-generating framework for epidermal stratification and experimental prioritization in AD.