Machine learning identifies cytokine signatures of disease severity and autoantibody profiles in systemic lupus erythematosus - a pilot study.

Disrupted cytokine networks and autoantibodies play an important role in the pathogenesis of systemic lupus erythematosus. However, conflicting reports and non-reproducibility have hindered progress regarding the translational potential of cytokines in SLE. This study attempts to address the existing knowledge gap using multiplex cytokine assay and machine learning. 67 SLE patients fulfilling SLICC criteria were recruited after informed consent, and circulating cytokines were measured by multiplex cytokine assay kit. We observed a positive association between actual disease activity scores (SLEDAI) and predicted scores from a partial least squares regression (PLSR) analysis of multivariate cytokine response data, with MIP-1α having a strong contribution towards disease activity. Our analysis also highlights increased IL-12 as a potential biomarker in nephritis and elevated MIP-1α as a signature of NPSLE. Using a k-Modes clustering algorithm to stratify patients based on patterns of co-occurrence of circulating autoantibodies, we identified 4 distinct clusters of patients. All 4 clusters had patients with nephritis, but patients in cluster 3 with nephritis were characterised by low levels of housekeeping and homeostatic cytokines and the presence of anti-Ro antibodies, which is a novel observation. Thus, we demonstrate that cytokines can be a surrogate to predict disease activity and organ involvement in SLE. Moreover, we show the utility of unsupervised machine learning algorithms using specific autoantibody signatures to predict renal involvement in SLE.