Many atopic dermatitis (AD) patients have suboptimal responses to Dupilumab therapy. This study identified key genes linked to this resistance using multi-omics approaches to benefit more patients. We selected a prospective cohort of 54 AD treated with Dupilumab from the GEO database. After identifying resistant genes via WGCNA and differential expression analysis, we used machine learning techniques to screen key genes and develop a predictive model. It was found that four key genes (AP2M1, BMP4, DNM1, and RHEB) were identified, showing excellent diagnostic performance for Dupilumab resistance (AUC = 0.832-0.861, P <
0.05) and validated in AD patients via RT-qPCR (P <
0.05). Among them, AP2M1 was significantly correlated with the clinical severity of AD (R = 0.5,P = 0.04) and identified as a potential risk factor (HR = 13.45, 95%CI(1.71-105.65), P = 0.02). The results of immunohistochemistry also revealed overexpression of AP2M1 in AD tissue (P = 0.002). Additionally, immune infiltration analysis suggested that AP2M1-mediated Dupilumab resistance may involve mast cells (R = -0.51, P = 0.02), which also supported by single-cell analysis. And we constructed a regulatory network of AP2M1. Finally, we explored the drug Fostamatinib, targeting AP2M1. In conclusion, AP2M1 may serve as a biomarker for those AD patients exhibiting suboptimal responses to Dupilumab.