Clustering Subjects in Genetic Studies with Self Organizing Maps

Several machine learning techniques have been applied for finding multi-loci associations among Single Nucleotide Polymorphisms (SNPs) and a disease. In this paper it is investigated whether Self Organizing Maps (SOMs) can generate clusters associated with a disease based on the genetic patterns of subjects. A batch categorical SOM that can handle missing data was used on Genome Wide Association (GWA) data on Multiple Sclerosis (MS). The association of the clusters generated with the disease were initially tested using the Pearson's chi square test and then the weights of the top clusters were used for investigating for SNP patterns. The results of the analyses reveal statistically significant associations between the generated clusters and the disease, indicating that SOMs can be used for multi-loci associations.