Multiple sclerosis (MS) is a relatively common neurodegenerative illness that frequently causes a large level of disability in patients. While its cause is not fully understood, it is likely due to a combination of genetic and environmental factors. Diagnosis of multiple sclerosis through a simple clinical examination might be challenging as the evolution of the illness varies significantly from patient to patient, with some patients experiencing long periods of remission. In this regard, having a quick and inexpensive tool to help identify the illness, such as DNA CpG (cytosine-phosphate-guanine) methylation, might be useful. In this paper, a technique is presented, based on the concept of Shannon Entropy, to select CpGs as inputs for non-linear classification algorithms. It will be shown that this approach generates accurate classifications that are a statistically significant improvement over using all the data available or randomly selecting the same number of CpGs. The analysis controlled for factors such as age, gender and smoking status of the patient. This approach managed to reduce the number of CpGs used while at the same time significantly increasing the accuracy.
Keywords: DNA methylation; entropy; multiple sclerosis.