Integrated analysis of multiple microarray studies to establish differential diagnostic models of Crohn’s disease and ulcerative colitis based on a metalloproteinase-associated module

BACKGROUND: The ulcerative colitis (UC) and Crohn’s disease (CD) subtypes of inflammatory bowel disease (IBD) are autoimmune diseases influenced by multiple complex factors. The clinical treatment strategies for UC and CD often differ, indicating the importance of improving their discrimination. METHODS: Two methods, robust rank aggregation (RRA) analysis and merging and intersection, were applied to integrate data from multiple IBD cohorts, and the identified differentially expressed genes (DEGs) were used to establish a protein−protein interaction (PPI) network. Molecular complex detection (MCODE) was used to identify important gene sets. Two differential diagnostic models to distinguish CD and UC were established via a least absolute shrinkage and selection operator (LASSO) logistic regression, and model evaluation was performed in both the training and testing groups, including receiver operating characteristic (ROC) curves, calibration plots and decision curve analysis (DCA). The potential value of MMP-associated genes was further verified using different IBD cohorts and clinical samples. RESULTS: Four datasets (GSE75214, GSE10616, GSE36807, and GSE9686) were included in the analysis. Both data integration methods indicated that the activation of the MMP-associated module was significantly elevated in UC. Two LASSO models based on continuous variable (Model_1) and binary variable (Model_2) MMP-associated genes were established to discriminate CD and UC. The results showed that Model_1 exhibited good discrimination in the training and testing groups. The calibration analysis and DCA showed that Model_1 exhibited good performance in the training group but failed in the testing group. Model_2 exhibited good discrimination, calibration and DCA results in the training and testing groups and exhibited greater diagnostic value. The effects of Model_1 and Model_2 were further verified in a new IBD cohort of GSE179285. The MMP genes exhibited high value as biomarkers for the discrimination of IBD patients using published cohort and immunohistochemistry (IHC) staining data. The MMP-associated gene levels were statistically significantly positively correlated with the levels of the differentially expressed cell types, indicating their potential value in differential diagnosis. The single-cell analysis confirmed that the expression of ANXA1 in UC was higher than that in CD. CONCLUSION: MMP-associated modules are the main differential gene sets between CD and UC. The established Model_2 overcomes batch differences and has good clinical applicability. Subsequent in-depth research investigating how MMPs are involved in the development of different IBD subtypes is necessary.