Immunological Profiling of Paediatric Inflammatory Bowel Disease Using Unsupervised Machine Learning

Abstract

Objectives:The current classification of inflammatory bowel disease (IBD) is based on clinical phenotypes, which is blind to the molecular basis of the disease. The aim of this study was to stratify a treatment-naïve paediatric IBD cohort through specific innate immunity pathway profiling and application of unsupervised machine learning (UML).Methods:In order to test the molecular integrity of biological pathways implicated in IBD, innate immune responses were assessed at diagnosis in 22 paediatric patients and 10 age-matched controls. Peripheral blood mononuclear cells (PBMCs) were selectively stimulated for assessing the functionality of upstream activation receptors including NOD2, toll-like receptor (TLR) 1-2 and TLR4, and the downstream cytokine responses (IL-10, IL-1β, IL-6, and TNF-α) using multiplex assays. Cytokine data generated were subjected to hierarchical clustering to assess for patient stratification.Results:Combined immune responses in patients across 12 effector responses were significantly reduced compared with controls (P = 0.003) and driven primarily by "hypofunctional" TLR responses (P values 0.045, 0.010, and 0.018 for TLR4-mediated IL-10, IL-1β, and TNF-α, respectively; 0.018 and 0.015 for TLR1-2 -mediated IL-10 and IL-1β). Hierarchical clustering generated 3 distinct clusters of patients and a fourth group of "unclustered" individuals. No relationship was observed between the observed immune clusters and the clinical disease phenotype.Conclusions:Although a clinically useful outcome was not observed through hierarchical clustering, our study provides a rationale for using an UML approach to stratify patients. The study also highlights the predominance of hypo-inflammatory innate immune responses as a key mechanism in the pathogenesis of IBD.