Download PDFOpen PDF in browserAutomatic Detection of Multiple Sclerosis Using Genomic ExpressionEasyChair Preprint 1072114 pages•Date: August 15, 2023AbstractMultiple sclerosis (MS) is the prevailing demyelinating disease and the leading cause of neurological disability in the young adult population. Recent microarray gene expression profiling studies have identified a number of genetic variants that contribute to the complex etiology of multiple sclerosis (MS). This study presents a comprehensive analysis of multiple sclerosis (MS) data using microarray technology and machine learning approaches. The goal was to develop a blood biomarker prediction model for the diagnosis of MS. Two experiments were conducted: the first involving Principal Component Analysis (PCA) as a dimension reduction method, and the second utilizing various feature selection techniques.in addition, online STRING database was used for prediction the gene interaction and functional annotation. In the first experiment, PCA was employed with LDA, SVM, and KNN classifiers optimized with different kernel functions. The best accuracy achieved was 95.83% with LDA using 26 components. SVM and KNN classifiers yielded accuracies of 91.67% and 87.5%, respectively. The second experiment focused on feature selection methods (Fisher score, chi-square, relief, and MRMR) combined with LDA, SVM, and KNN classifiers. The best results were obtained with the relief feature selection method, achieving 100% accuracy with KNN using 38 DEG. Fisher score, chi-square and MRMR methods showed higher accuracies of 91.6% ,87.5 and 87.5%, respectively. Functional annotation indicates that these 38 DEG associated with immune and neurological functions. Furthermore, the analysis result suggested that MIF, PTGES3, CYLD and JAK1 may play central roles in gene expression in the pathogenesis of MS. Keyphrases: Microarray Technology, blood biomarker, dimension reduction, machine learning
|