In this diploma thesis our goal was to classify viral sequences into taxonomic groups by using different machine learning methods. We assembled the taxonomic structure by collecting data from NCBI web site. To clean the data we applied several filtering steps. We then evaluated the predictive performance of classical and structured machine learning methods on the task of classification in taxonomy groups. We wanted to determine the most suitable way to describe genomic sequences. Using k-mers to describe the genomic composition yielded poor predictive models, with best performance slightly above the performance of the majority classifier. Methods, which are able to use prior knowledge on the taxonomic relationships between classes, performed slightly better than methods, which did not use such information.
|
