An AI-assisted morphoproteomic approach is a supportive tool in esophagitis-related precision medicine

Esophagitis is a frequent, but at the molecular level poorly characterized condition with diverse underlying etiologies and treatments. Correct diagnosis can be challenging due to partially overlapping histological features. By proteomic profiling of routine diagnostic FFPE biopsy specimens (n = 55) representing controls, Reflux- (GERD), Eosinophilic-(EoE), Crohn’s-(CD), Herpes simplex (HSV) and Candida (CA)-esophagitis by LC-MS/MS (DIA), we identified distinct signatures and functional networks (e.g. mitochondrial translation (EoE), immunoproteasome, complement and coagulations system (CD), ribosomal biogenesis (GERD)), and pathogen-specific proteins for HSV and CA. Moreover, combining these signatures with histological parameters in a machine learning model achieved high diagnostic accuracy (100% training set, 93.8% test set), and supported diagnostic decisions in borderline/challenging cases. Applied to a young patient representing a use case, the external GERD diagnosis could be revised to CD and ICAM1 was identified as highly abundant therapeutic target. This resulted in CyclosporinA as a personalized treatment recommendation by the local multidisciplinary molecular inflammation board. Our integrated AI-assisted morphoproteomic approach allows deeper insights in disease-specific molecular alterations and represents a promising tool in esophagitis-related precision medicine. Proteomic analysis of routine biopsy specimens of esophagitis patients revealed distinct networks. Combining proteomic data with histological parameters in a machine learning model achieved high diagnostic accuracy and supported diagnostic decisions in challenging cases. Identification of disease-specific proteomic profiles of relevant differential diagnoses of esophagitis (GERD, EoE, HSV, CA and CD) via LC-MS/MS using routine esophageal biopsies. Introducing a morpho-proteomic machine learning approach (random forest, RF) for tissue-based diagnostics of esophagitis on FFPE-Samples. The morpho-proteomic RF approach supports esophagitis diagnostics especially by (re-) classifying borderline and complex cases. Applicability in personalized medicine to identify therapeutic targets for tailored treatment recommendations and due to the use of FFPE-samples is transferable to virtually all diseases involving tissue-based diagnostics. Identification of disease-specific proteomic profiles of relevant differential diagnoses of esophagitis (GERD, EoE, HSV, CA and CD) via LC-MS/MS using routine esophageal biopsies. Introducing a morpho-proteomic machine learning approach (random forest, RF) for tissue-based diagnostics of esophagitis on FFPE-Samples. The morpho-proteomic RF approach supports esophagitis diagnostics especially by (re-) classifying borderline and complex cases. Applicability in personalized medicine to identify therapeutic targets for tailored treatment recommendations and due to the use of FFPE-samples is transferable to virtually all diseases involving tissue-based diagnostics. Proteomic analysis of routine biopsy specimens of esophagitis patients revealed distinct networks. Combining proteomic data with histological parameters in a machine learning model achieved high diagnostic accuracy and supported diagnostic decisions in challenging cases.