Machine learning encodes urine and serum metabolic patterns for autoimmune disease discrimination, classification and metabolic dysregulation analysis

• Published in: Analyst (London) Vol. 148; no. 18; pp. 4318 - 433
• Main Authors: Du, Qiuyao, Wang, Xiao, Chen, Junyu, Wang, Yiran, Liu, Wenlan, Wang, Liping, Liu, Huihui, Jiang, Lixia, Nie, Zongxiu
• Format: Journal Article
• Language: English
• Published: England Royal Society of Chemistry 11.09.2023
• Subjects: Algorithms; Autoimmune diseases; Chronic conditions; Classification; Connective tissues; Diagnosis; Disease; Liquid chromatography; Machine learning; Mass spectrometry; Metabolic disorders; Metabolism; Metabolites; Pathogenesis; Signs and symptoms; Urine
• ISSN: 0003-2654; 1364-5528
• DOI: 10.1039/d3an01051a

There is a wide variety of autoimmune diseases (ADs) with complex pathogenesis and their accurate diagnosis is difficult to achieve because of their vague symptoms. Metabolomics has been proven to be an efficient tool in the analysis of metabolic disorders to provide clues about the mechanism and diagnosis of diseases. Previous studies of the metabolomics analysis of ADs were not competent in their discrimination. Herein, a liquid chromatography tandem mass spectrometry (LC-MS) strategy combined with machine learning is proposed for the discrimination and classification of ADs. Urine and serum samples were collected from 267 subjects consisting of 127 healthy controls (HC) and 140 AD patients, including those with rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), sicca syndrome (SS), ankylosing spondylitis (AS), systemic scleroderma (SSc) and connective tissue disease (CTD). Machine learning algorithms were encoded for the discrimination and classification of ADs with metabolomic patterns obtained by LC-MS, and satisfactory results were achieved. Notably, urine samples exhibited higher accuracy for disease differentiation and triage than serum samples. Apart from that, differential metabolites were selected and metabolite panels were evaluated to demonstrate their representativeness. Metabolic dysregulations were also investigated to gain more knowledge about the pathogenesis of ADs. This research provides a promising method for the application of metabolomics combined with machine learning in precision medicine. Machine learning of urine and serum metabolic patterns encodes the discrimination and classification of autoimmune diseases. The selected metabolite panel, metabolite dysregulation and disturbance pathways were investigated.