A new study has identified a three-gene lysosomal signature that accurately distinguishes pediatric ulcerative colitis cases and helps estimate both disease severity and the risk of surgical colectomy. The investigation focused on three specific hub genes—caudal type homeobox 2, sialic acid acetylesterase, and myotubularin 1—which exhibited pronounced downregulation within pediatric disease lesions. By incorporating these molecular markers into machine learning frameworks, researchers demonstrated high predictive performance for clinical classification and patient trajectory assessment.
While cellular lysosomes are recognized to play critical roles in the broader pathogenesis of inflammatory bowel disease, the precise involvement of lysosome-related molecules within pediatric ulcerative colitis has remained poorly understood. To address this knowledge gap, the study sought to pinpoint key lysosome-associated hub genes and utilize them to build computational models capable of diagnostic discrimination and prognostic risk prediction in affected children.
To establish candidate biomarkers, the investigators obtained transcriptomic datasets from the public Gene Expression Omnibus database. They filtered the candidate hub genes through a combination of differential expression profiling and weighted gene co-expression network analysis.
The experimental analyses confirmed that the three hub genes, known formally as CDX2, SIAE, and MTM1, were markedly downregulated in pediatric ulcerative colitis lesions. Importantly, reduced expression levels of these three genes demonstrated significant negative correlations with established clinical indices, specifically the Pediatric Ulcerative Colitis Activity Index score and the total Mayo score. Furthermore, tissue expression of these genes was observed to be notably lower in pediatric patients who ultimately required colectomy within one year of their initial diagnosis.
Using the expression profiles of these three genes, the researchers constructed four distinct diagnostic machine learning architectures: random forest, support vector machine, linear discriminant analysis, and k-nearest neighbor models. Across training cohorts, these diagnostic models achieved area under the curve metrics spanning from 0.990 to 0.998. When evaluated in independent test cohorts, the diagnostic area under the curve values ranged between 0.870 and 0.981, with the random forest and linear discriminant analysis algorithms demonstrating superior stability.
In diagnostic testing, the random forest model reached a test-cohort area under the curve of 0.981, while the linear discriminant analysis model attained 0.970. Building on these diagnostic results, the research team applied the random forest and linear discriminant analysis algorithms to predict the specific likelihood of colectomy within one year of diagnosis. For this surgical outcome, the random forest model achieved an area under the curve of 0.932 in training and 0.905 in testing.
The linear discriminant analysis model similarly exhibited strong prognostic utility for colectomy prediction, recording an area under the curve of 0.850 in the training cohort and 0.930 in the test cohort. Beyond classification performance, biological correlation analyses revealed that CDX2, SIAE, and MTM1 expression was positively linked with genes maintaining intestinal epithelial barrier integrity. Conversely, the expression of these three genes was negatively correlated with genes associated with intestinal inflammation and fibrotic tissue remodeling.
The study authors concluded that the lysosome-related hub genes CDX2, SIAE, and MTM1 are intimately connected to the pathophysiology of pediatric ulcerative colitis. According to the research findings, this three-gene panel holds utility for assisting in diagnostic confirmation, evaluating clinical disease severity, and indicating patient vulnerability to rapid colectomy. These observations provide fresh insights into the molecular landscape of the condition and offer potential quantitative markers to support future clinical management strategies.
Subject of Research: Machine learning evaluation of lysosome-related genes in pediatric ulcerative colitis diagnosis and colectomy risk
Article Title: Machine learning-based lysosome-related gene signature for diagnosis, severity stratification, and colectomy risk prediction in pediatric ulcerative colitis
Article References: Wu, B., Huang, L., Yang, S., Liu, J., Lv, Q., Xu, Q., & Shao, M. (2026). Machine learning-based lysosome-related gene signature for diagnosis, severity stratification, and colectomy risk prediction in pediatric ulcerative colitis. BMC Gastroenterology. https://doi.org/10.1186/s12876-026-05616-5
Image Credits: AI Generated
DOI: 10.1186/s12876-026-05616-5
Keywords: pediatric ulcerative colitis, ulcerative colitis, inflammatory bowel disease, lysosome-related genes, CDX2, SIAE, MTM1, machine learning, random forest, linear discriminant analysis, support vector machine, k-nearest neighbor, colectomy risk, PUCAI score, Mayo score, intestinal epithelial barrier, intestinal fibrosis, qRT-PCR validation
News Source: Harold Sullivan. (November 20, 2026). Three Lysosome Genes Predict Pediatric Ulcerative Colitis and Colectomy Risk. Scienmag.



