Characterizing Duodenal Immune Microenvironment in Functional Dyspepsia: An AutoML-Driven Diagnostic Framework

Functional dyspepsia (FD) is a prevalent gastroduodenal disorder with an unclear pathogenesis. Recent studies suggest that duodenal immune activation plays a pivotal role in its development.

Mendelian randomization analysis using genome-wide association studies (GWAS) and expression quantitative trait loci (eQTL) data identified genes associated with FD. Expression data from 40 FD patients and 24 healthy controls were analyzed for differentially expressed genes (DEGs) using the Gene Expression Omnibus (GEO) database. Immune infiltration was assessed using CIBERSORT and xCell algorithms, followed by weighted gene co-expression network analysis (WGCNA) to identify immune-related gene modules. The top 20 critical genes were selected using maximal clique centrality (MCC), and a diagnostic model was developed using LASSO regression and multivariate logistic regression. We utilized the AutoGluon framework to automate the construction and optimization of a model based on hub genes, generating a high-performance diagnostic model. GO/KEGG and GSEA analyses were utilized to further explore the potential biological mechanisms of hub genes. Finally, FD rat models were validated for central gene and immune cell expression.

GWAS identified 259 genes causally linked to FD, enriched in immune and inflammatory processes. Expression analysis showed altered immune infiltration, with increased plasma cells and decreased regulatory T cells (p < 0.05). Nine biomarkers (AMBP, CHGA, GCG, SOX9, TTR, CCK, CLU, RBP4, SST) showed excellent diagnostic performance (AUC=0.94). The AutoGluon framework identified LightGBMLarge and XGBoost models as the best for clinical application. GSEA revealed high-risk groups were linked to immune-inflammatory and cell adhesion pathways. GO/KEGG analysis associated high-risk scores with oxidative stress, immune response, and nutrient metabolism.

This study proposes immune-related diagnostic biomarkers in FD, offering insights into FD pathogenesis and potential therapeutic targets. The automated machine learning model accurately identifies high-risk FD patients and characterizes biological changes, providing new perspectives for understanding FD.