AB0155 Learning sle pathological mechanisms from multi 'omics profiles. (15th June 2017)

• Record Type: Journal Article
• Authors: Pfister, SS; Kelley, N; Schlitt, T; Fernandez, A; Sultan, M; Hasan, M; Mir, A; Rush, JS
• Is Part Of: Annals of the rheumatic diseases. Volume 76(2017)Supplement 2
• Page Start: 1101
• Page End: 1101
• Publication Date: 2017-06-15
• Subjects: Rheumatism -- Periodicals; 616.723005
• DOI: 10.1136/annrheumdis-2017-eular.5009
• Languages: English
• ISSNs: 0003-4967

Abstract

Abstract : Background: Precision medicine aims at providing intervention based on clinical and molecular stratification of patients, and is an important approach for targeting heterogeneous diseases. A diverse autoimmune disease is systemic lupus erythematosus (SLE), where dysregulation of several immune processes affects multiple organs. Fundamental for targeted treatment of such a heterogeneous disease is the identification of biomarkers predictive for the biological basis of clinical phenotypes. Objectives: Despite recent progress, few markers for SLE are currently used in the clinic. In order to learn SLE pathological mechanisms and associated biomarkers, we obtained a diverse dataset from a cohort of active SLE patients (SLEDAI >6), including blood transcriptomics, serum proteomics, cytokines, and auto-antibody profiles. Integration of multi-omics data provides a rich dataset to explore associations between molecular and clinical readouts. Methods: From a machine-learning perspective, biomarker discovery is defined as the process of selecting an optimal subset of variables for the prediction of parameters of interest. However, variable selection approaches are often underpowered for datasets that contain fewer samples than the number of variables. To overcome this problem we present a method based on L1 regularized regression and recursive variable elimination to generate networks of predictive markers across multiple data types. Results: The proposed method allows us to graphically visualize the relationships among SLE phenotypes, and their molecular fingerprints. Identified networks of markers are validated by mapping to known biological pathways, and when available by comparison to independent patient cohorts. Despite the small number of patients (n=20), we identify known pathological mechanisms, including a type I IFN gene signature, several cell type specific signatures, and potential novel markers of clinically defined SLE subtypes. Conclusions: Systemic lupus erythematosus is a complex autoimmune disease characterized by a variety of clinical manifestations. While multi-omics profiles from SLE patients pose challenges because of their intrinsic high dimensionality, they also provide a unique insight into the molecular processes of disease. Our integrated analysis gives a novel perspective on the pathological mechanisms of clinical SLE phenotypes. Disclosure of Interest: None declared