GPX8, COL1A1, RAB31, and Other Genes as Diagnostic Biomarkers for Crohn's Disease: A Bioinformatics Analysis Using...

To identify potential pivotal genes associated with Crohn's disease (CD) that may serve as potential diagnostic biomarkers.

A bioinformatics analysis. Place and Duration of the Study: Department of Gastrointestinal Dysfunction Centre, Anhui No.2 Provincial People's Hospital, Hefei, China, from June 2023 to June 2024.

Multiple gene expression datasets from the NCBI Gene Expression Omnibus (GEO) were used to identify potential diagnostic biomarkers. Weighted gene co-expression network analysis (WGCNA) was conducted; CD-related genes were selected through WGCNA, gene-gene interaction network analysis, and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Logistic regression and six supervised learning techniques-including support vector machine, random forest, K-nearest neighbour, neural network, decision tree, and extreme gradient boosting (XGB)-were used to assess and compare the diagnostic performance of various models.

Hub gene screening, immune infiltration analysis, and gene set enrichment analysis were used to explore the association between the potential genes and CD. A nomogram based on the logistic regression model was developed, and the clinical utility of the hub genes was evaluated through decision curve analysis (DCA). Additionally, six machine learning models were developed, with the XGB model demonstrating the highest performance. The area under the curve (AUC) for the XGB model was 99.4% for the training set, 77.7% for the validation set, and 83.5% for external validation data, indicating its superior diagnostic potential.

This study identified five genes that are closely associated with CD. Based on these genes, a CD diagnostic model was established, with the XGB model showing the most promising results among the six tested models.

Crohn's disease, Genes, Prediction model, Machine learning.