3003-LB: Machine Learning–Based Clustering of Genetic Risk Profiles Reveals Distinct Endotypes of Type 1 Diabetes

Introduction and Objective: Improved definition of endotypes for type 1 diabetes (T1D) can facilitate the development of targeted therapies and improve clinical management of disease. We recently discovered four distinct sub-groups of T1D derived from a machine learning-based genetic risk score, which had significant differences in clinical parameters such as T1D age of onset and rate of T1D complications. Here we further expanded on the characterization and evaluation of these sub-groups as potential endotypes of T1D. Methods: We mapped 29,746 T1D and non-diabetic individuals of European Ancestry to four genetic sub-groups using our machine learning method, and performed association testing, allele frequency enrichment, gene and pathway enrichment and risk scoring for related autoimmune diseases for each sub-group. Results: Genetic sub-clusters demonstrated biologically and clinically significant differences in genetic features and autoimmune predisposition (Figure 1), which supports that they represent endotypes of T1D. Conclusion: Sub-grouping of individuals based on genetic risk profiles revealed endotypes of T1D with distinct clinical and mechanistic properties. Analyses in patient cohorts will further validate the clinical and therapeutic value of these endotypes. Disclosure A.L. Ghaben: None. E. Griffin: None. K. Gaulton: Other - Spouse employee; Current; Altos Labs. Stock/Shareholder; Current; Neurocrine Biosciences, Inc.