김보경 (울산의과대학교)

김경곤 (울산의과대학교)

김보경 (울산의과대학교)

안보경 (서울아산병원)

고창석 (서울아산병원)

김지현 (울산의과대학교)

이예린 (울산의과대학교)

김민중 (울산의과대학교)

유지영 (서울아산병원)

김범수 (서울아산병원)

홍승모 (서울아산병원)

김경곤 (울산의과대학교)

Inflammation within the tumor microenvironment is increasingly recognized as a critical determinant of prognosis and therapeutic response across diverse cancers, including gastric neuroendocrine carcinomas (GNECs). However, the proteomic features that differentiate inflamed from non-inflamed GNECs remain poorly characterized. In this study, we performed LC-MS/MS-based spatial proteomic profiling on 30 tumor tissue slides from 20 GNEC patients, stratified into inflamed (n = 20) and non-inflamed (n = 10) groups based on histopathological evaluation. Protein quantification was conducted using label-free data-dependent acquisition on the Vanquish Neo system coupled with an Orbitrap Exploris™ 480 mass spectrometer. To identify inflammation-associated protein signatures, we applied conventional machine learning methods such as Support Vector Machine (SVM), as well as ensemble models including Random Forest (RF) and XGBoost (XGB). Performance of protein panel was assessed using recall, F1 score, precision, accuracy, and AUC-ROC. Model robustness was further validated in an independent patient cohort. Feature importance analysis revealed a distinct set of proteins most predictive of inflammatory status. Spatial clustering demonstrated clear separation between inflamed and non-inflamed groups based on molecular features, underscoring biological relevance. These findings underscore the value of integrating mass spectrometry-based proteomics with computational modeling to better understand tumor heterogeneity in GNECs. The identified protein signatures show potential as prognostic biomarkers and may support the development of personalized therapeutic strategies.

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. RS-2024-00454407).