1111-P: Integrating Longitudinal Clinical and Baseline Multiomics Data for Predicting C-Peptide Progression in Newly Diagnosed Type 1 Diabetes

Abstract

Background: Development of disease-modifying therapy for type 1 diabetes (T1D) is hampered by the limited understanding of pathogenesis, heterogeneity, lack of disease biomarkers and stratifiers. Using the INNODIA consortium pan-European infrastructure to collect prospective clinical data from newly diagnosed subjects combined with multi-omics data, we mined integrated datasets using deep learning to identify novel relationships that could transform the disease monitoring landscape. Method: Samples collected <6 weeks from diagnosis from 100 islet autoantibody positive participants were prioritised for omics analysis and the data integrated with clinical parameters and C-peptide (CP) from mixed meal tolerance tests (at 3, 6 and 12 months). SNP array data, plasma metabolomics, lipidomics and miRNAs, serum proteomics, whole blood RNA-sequencing and multiFACS immunomics from PBMCs were collected. We applied a variational autoencoder (VAE) to simultaneously integrate the data and learn CP progression over 12 months. The latent (lower dimensional) representation of the VAE integrated data was clustered to identify clinically meaningful subgroups. Results: Clustering identified five age-independent subgroups with significantly different progression patterns (p<0.05). Cluster 1 and 3 associated with lower fasted and stimulated CP. Cluster 2 and 5 associated with higher fasted CP, while cluster 4 most strongly associated with higher stimulated CP. Cluster 1 had the lowest stimulated CP at baseline, cluster 5 the lowest fasted CP and cluster 4 the highest fasted and stimulated CP at baseline. Investigation of the most predictive signals from the omics data is ongoing. Discussion: The preliminary results suggest that the VAE can learn meaningful progression patterns from integrated clinical and multi-omics data, which holds great promise to uncover important data structures not readily discovered when analyzing single data types. Disclosure C. Brorsson: Employee; Spouse/Partner; Novo Nordisk A/S. J. Almagro armenteros: None. G. Mazzoni: Employee; Self; Novo Nordisk A/S. S. Kaur: None. A. M. Schulte: None. C. Mathieu: Advisory Panel; Self; Novo Nordisk, Sanofi, Merck Sharp and Dohme Ltd., Eli Lilly and Company, Novartis, AstraZeneca, Boehringer Ingelheim, Roche, Medtronic, ActoBio Therapeutics, Pfizer, Insulet and Zealand Pharma, Research Support; Self; Medtronic, Novo Nordisk, Sanofi and ActoBio Therapeutics, Speaker’s Bureau; Self; Novo Nordisk, Sanofi, Eli Lilly and Company, Boehringer Ingelheim, AstraZeneca and Novartis. S. Brunak: Board Member; Self; Intomics A/S, Proscion A/S, Stock/Shareholder; Self; Hoba Therapeutics Aps, Novo Nordisk A/S. On behalf of the innodia consortium: n/a. Funding INNODIA (115797)