The feasibility of developing biomarkers from peripheral blood mononuclear cell RNAseq data in children with juvenile idiopathic arthritis using machine learning approaches

Abstract

Background: The response to treatment for juvenile idiopathic arthritis (JIA) can be staged using clinical features. However, objective laboratory biomarkers of remission are still lacking. In this study, we used machine learning to predict JIA activity from transcriptomes from peripheral blood mononuclear cells (PBMCs). We included samples from children with Native American ancestry to determine whether the model maintained validity in an ethnically heterogeneous population. Methods: Our dataset consisted of 50 samples, 23 from children in remission and 27 from children with an active disease on therapy. Nine of these samples were from children with mixed European/Native American ancestry. We used 4 different machine learning methods to create predictive models in 2 populations: The whole dataset and then the samples from children with exclusively European ancestry. Results: In both populations, models were able to predict JIA status well, with training accuracies > 74% and testing accuracies > 78%. Performance was better in the whole dataset model. We note a high degree of overlap between genes identified in both populations. Using ingenuity pathway analysis, genes from the whole dataset associated with cell-to-cell signaling and interactions, cell morphology, organismal injury and abnormalities, and protein synthesis. Conclusions: This study demonstrates it is feasible to use machine learning in conjunction with RNA sequencing of PBMCs to predict JIA stage. Thus, developing objective biomarkers from easy to obtain clinical samples remains an achievable goal.