Metabolomic and machine learning enhances patient diagnosis and stratification in systemic autoimmune diseases

Abstract

Systemic Autoimmune Diseases (SADs) present clinical challenges due to their heterogeneity, which complicates patient classification, and delays diagnosis. We characterized their metabolomic fingerprints aiming to uncover novel molecular insights and enhance patient stratification and diagnosis through the application of Machine Learning (ML). A total of 716 individuals from the international multicenter study PRECISESADS were included: 272 with Rheumatoid Arthritis (RA), 183 with Systemic Lupus Erythematosus (SLE), 148 with Antiphospholipid Syndrome (APS), 70 with Systemic Sclerosis (SSc), and 43 Healthy Donors (HDs). The circulating metabolomic profile was analyzed using Nuclear Magnetic Resonance (NMR) spectroscopy and a combination of supervised and unsupervised ML methods. Several metabolites were differentially expressed in each disease compared to HDs, with the highest number of alterations observed in SSc (99) and APS (68), followed by SLE (30) and RA (17). The prominent reduction of antioxidant and anti-inflammatory metabolites (albumin and histidine), combined with the increase in the pro-inflammatory marker GlycA, emerged as key shared hallmarks of SADs. Each disease also displayed a distinct set of uniquely altered metabolites. ML demonstrated strong diagnostic potential (AUC 0.79–0.87) by generating disease-specific signatures driven by alterations in lipids, fatty acids, energy metabolism, and amino acid pathways. Unsupervised clustering analysis of the entire cohort identified three distinct clusters, with each disease represented across all clusters in varying proportions, which were strongly associated with distinct key clinical features. This study highlights the utility of metabolomics and ML to classify and stratify patients with SADs, reinforcing their clinical relevance in precision medicine.