Combining S100B and Cytokines as Neuro-Inflammatory Biomarkers for Diagnosing Generalized Anxiety Disorder: A Proof-of-Concept Study Based on Machine Learning

Source abstract

Introduction S100 calcium-binding protein B (S100B) is a neurotrophic factor that regulates neuronal growth and plasticity by activating astrocytes and microglia through the production of cytokines involved in Generalized Anxiety Disorder (GAD). However, few studies have combined S100B and cytokines to explore their role as neuro-inflammatory biomarkers in GAD. Methods Serum S100B and cytokines (IL-1β, IL-2, IL-4, and IL-10) of 108 untreated GAD cases and 123 healthy controls (HC) were determined by enzyme-linked immunosorbent assay (ELISA), while Hamilton Anxiety Rating Scale (HAMA) scores and Hamilton Depression Rating Scale (HAMD) scores were measured to evaluate anxiety and depression severity. This was used to help physicians identify persons having GAD. Machine learning techniques were applied for feature ordering of cytokines and S100B and the classification of persons with GAD and HC. Results The serum S100B, IL-1β, and IL-2 levels of GAD cases were significantly lower than HC ( P < 0.001), and the IL-4 level in persons with GAD was significantly higher than HC ( P < 0.001). At the same time, IL-10 had no significant difference between the two groups ( P = 0.215). The feature ranking distinguishing GAD from HC using machine learning ranked the features in the following order: IL-2, IL-1β, IL-4, S100B, and IL-10. The accuracy of S100B combined with IL-1β, IL-2, IL-4, and IL-10 in distinguishing persons with GAD from HC was 94.47 ± 2.06% using an integrated back propagation neural network based on a bagging algorithm (BPNN-Bagging). Conclusion The serum S-100B, IL-1β, and IL-2 levels in persons with GAD were down-regulated while IL-4 was up-regulated. The combination of S100B and cytokines had a good diagnosis value in determining GAD with an accuracy of 94.47%. Machine learning was a very effective method to study neuro-inflammatory biomarkers interacting with each other and mediated by plenty of factors.