Exploiting Machine Learning And Gene Expression Analysis in Amyotrophic Lateral Sclerosis Diagnosis

Despite many research efforts, the biological insight related to Amyotrophic Lateral Sclerosis (ALS), a rare disease resulting in the loss of motor neurons and causing mortality, remains elusive and leads to challenges to the diagnosis of the disease. Fortunately, gene expression data has recently appeared as a potential approach for the functionality analysis of genes related to orphan diseases and for providing more accurate diagnosis outcomes. Moreover, with the explosion of machine learning (ML), implementing ML in analyzing biomedical data has become a promising direction with a notable effect on our lives. Leveraging these advantages, in this paper, we investigate to shed light on the effects of gene markers on ALS diagnosis and propose a novel gene combination that is effective in ALS diagnosis. We retrieve the datasets and perform the cleaning and pre-processing methods to obtain robust data for analysis. Then, the Max-Min Parents and Children (MMPC) and Sequential Forward Feature Selection (SFFS) algorithms are applied to achieve the optimal gene subsets that are effective for the final intelligent diagnosis model. Notably, the coefficient of the Ridge Classifier is utilized as the crucial score for determining the gene importance ranking table based on the selected gene signatures. All the possible gene combinations are evaluated and optimized in a set of robust machine learning algorithms. Consequently, a set of 20 genes identified through the Support Vector Machine (SVM) algorithm is selected as the optimal for the ALS diagnosis with an accuracy of 88.30% and an AUC score of 91.11%, which is dominant in comparison with notable traditional methods under the same datasets.