ABSTRACT
Sjögren's syndrome is an autoimmune disease that can also occur in children. The disease is not well defined and there is limited information on the presence of chemokines, cytokines, and biomarkers (CCBMs) in the saliva of children that could improve their disease diagnosis. In a recent study [1], we reported a large dataset of 105 CCBMs that were associated with both lymphocyte and mononuclear cell functions [2] in the saliva of 11 children formally diagnosed with Sjögren's syndrome and 16 normal healthy children. Here, we extend those findings and use the Mendeley dataset [2] to identify CCBMs that have predictive power for Sjögren's syndrome in female children. Datasets of CCBMs from all saliva samples and female children saliva samples were standardized. We used machine learning methods to select Sjögren's syndrome associated CCBMs and assessed the predictive power of selected CCBMs in these two datasets using receiver operating characteristic (ROC) curves and associated areas under curve (AUC) as metrics. We used eight classifiers to identify 16 datasets that contained from 2 to 34 CCBMs with AUC values ranging from 0.91 to 0.94.
