Unveiling potential biomarkers for systemic lupus erythematosus and lupus nephritis through metabolomics and machine learning

Systemic lupus erythematosus (SLE) is a common autoimmune disorder with increasing global incidence. Lupus nephritis (LN), a severe complication of SLE, occurs in approximately 50% of patients and is a leading cause of morbidity and mortality. This study aims to identify novel diagnostic biomarkers to facilitate the diagnosis of SLE and LN. Using non-targeted metabolomics, we analyzed plasma samples from healthy controls (HC), SLE patients and LN patients to identify differentially expressed metabolites. Machine learning approaches were applied to evaluate the diagnostic potential of these metabolites alone or in combination with clinical indicators. We identified 374, 528 and 400 differentially expressed metabolites in SLE vs. HC, LN vs. HC and LN vs. SLE, respectively. Among these, 3-Hydroxymethyl-Picumast showed the most pronounced decreasing trend in LN vs. SLE vs. HC. ROC analysis revealed that the combination of 3-Hydroxymethyl-Picumast, C3 and 3Beta, 7Alpha-Dihydroxy-5-Cholestenoate yielded the highest diagnostic accuracy (AUC = 72.92%). C3 alone exhibited the highest sensitivity (100%). 3Beta, 7Alpha-Dihydroxy-5-Cholestenoate showed the highest specificity (97.36%). Our findings suggest that 3-Hydroxymethyl-Picumast, particularly in combination with C3 and 3Beta, 7Alpha-Dihydroxy-5-Cholestenoate, holds promise as a diagnostic biomarker for distinguishing LN from SLE. This approach may enhance diagnostic evaluation, offering a theoretical foundation for improved clinical management.