Identification of Novel Biomarkers for Crohn's Disease Through the Integration of Machine Learning, Colocalization, and SMR Analysis

Abstract

Crohn's disease (CD) is a chronic inflammatory bowel disease with a prevalence rate increasing with time, thus demanding improved diagnostic and therapeutic strategies. The present work focused on identifying the candidate biomarkers for CD diagnosis and treatment. Gene Expression Omnibus (GEO)-derived CD-related gene expression datasets were analyzed. Differential protein–protein interaction network and weighted gene co-expression network analyses were conducted to prioritize the core candidate genes. Multiple machine learning algorithms were used to further refine these candidates. The feature importance of the model with the highest performance was explained using SHapley Additive exPlanations. Additionally, a single-sample gene set enrichment analysis was carried out to evaluate immune cell infiltration and determine the associations with diagnostic markers. In addition, the causal biomarker genes were identified using Bayesian colocalization and the summary data-based Mendelian randomization (SMR) analysis. The combination of glmBoost and random forest machine learning analysis identified five hub genes (CXCL5, SERPINB2, SOCS3, PF4, and IL1R1), which demonstrated robust diagnostic performance for CD. These biomarkers were correlated with the immune cell infiltration patterns indicative of heightened inflammation and Th1/Th17 adaptive immune responses. Colocalization and SMR analyses established a causal association of IL1R1 with CD development. This integrative multiomics approach identified the key biomarkers involved in the pathogenic mechanism of CD. The eQTL data based SMR analysis suggested a significant association of IL1R1 with CD risk, highlighting its dual effects as a diagnostic biomarker and therapeutic target.