Machine learning in the identification of phenotypes of multiple sclerosis patients

Multiple sclerosis (MS) is a complex disease affecting the central nervous system, mainly in young adults. Even though there is no cure for this disease, correct phenotype identification on time will help the clinicians to select proper treatment option, which can delay the severity progression and the long-term disability progression of the disease. As automated phenotype identification system can improve the situation in a positive way, this study explores the role of exon level transcriptome data in the prediction of three different phenotypes of this disease: subgroups, therapy option for individual patient and duration of the disease. Features selected from the downloaded transcriptome data using feature importance and mutual information are used along with different machine learning models to choose the best one with the highest accuracy. Comparison between models shows that while classifying the patients into their corresponding subgroups this study achieves the best accuracy value of 0.88 ± 0.05. In the prediction of therapy option our accuracy is 0.94 ± 0.07and in disease duration prediction the root mean square value is 4.1 ± 1.4. Notably, all the highest accuracies obtained here are using same machine learning algorithms and the value shows that they can be used in clinical practices.