ORLANDO—RNA expression patterns are highly dynamic, even at the earliest stages of multiple sclerosis (MS) pathogenesis, according to research presented at the ACTRIMS 2017 Forum. Machine learning methods trained and validated with these gene targets are a highly accurate tool that could provide actionable data for health care providers who suspect MS.
Charles F. Spurlock III, PhD
RNA sequencing was performed to identify differentially expressed protein-coding and noncoding RNA genes at distinct stages of MS, compared with controls. Expression levels of these genes were validated by real-time polymerase chain reaction (PCR) across all 979 subjects recruited. Ratios of gene expression data were used as inputs to train machine learning classifiers capable of multicategory comparisons. An independent validation testing set consisting of individuals across each control and disease class was used to evaluate the performance of these classifiers. In contrast to protein-coding gene targets where gene expression differences were often twofold or less, lncRNAs were highly differentially expressed across each of the experimental groups compared. Training and validation of machine learning methods revealed that lncRNA gene expression inputs resulted in higher overall accuracy and confidence of machine learning predictions than did protein-coding genes, resulting in greater than 90% accuracy in the validation testing set.
—Glenn S. Williams