Clinical classification of multiple sclerosis by machine learning

Multiple sclerosis (MS) is an autoimmune demyelinating disease of the central nervous system that affects many adults and causes substantial neurological disability. Its clinical presentations are highly heterogeneous, and distinguishing clinically isolated syndrome (CIS), relapsing-remitting MS and progressive subtypes remains difficult, even though subtype determines treatment. Machine learning offers a way to address this difficulty: it identifies subtle patterns in high-dimensional biomedical data and has shown promise for MS classification from neuroimaging, whereas its performance on transcriptomic data has received less attention. To examine this question, we benchmarked three ensemble classifiers, AdaBoost, Random Forest and XGBoost, on 594 peripheral-blood mononuclear-cell transcriptomes from three public datasets, using a shared 6,033-gene matrix for four binary tasks: MS versus non-MS, CIS versus non-CIS, primary progressive versus relapsing-remitting MS, and secondary progressive versus relapsing-remitting MS. Models were evaluated on a 3:1 training/held-out split after standardization and two-component principal component analysis. All three classifiers achieved high held-out performance across the four tasks. These results indicate that ensemble machine learning can effectively classify MS and its clinical subtypes from blood transcriptomic profiles.