Metodologias de aprendizagem automática na descoberta de interações entre genes em doenças complexas

Abstract

Gene-gene interactions have been found to involve a considerable amount of SNPs. However, many Genome Wide Association Studies (GWAS) have been performed with interesting results. The amount of computing power required was not possible to achieve until recently. Machine Learning methodologies quickly became a successful tool to find previously unknown genetic relations. The popularity of this field increased greatly after discovering the potential value of epistatic studies in the detection and understanding of how phenotypes are expressed. The purpose of this dissertation is to increase the knowledge obtained from these studies, detecting more interactions related to the manifestation of complex diseases.To achieve this goal, there are two main stages. The first stage consists of a comparison study amongst the most recent statistical and machine learning methodologies using simulated data sets containing generated epistatic interactions. The algorithms BEAM3.0, BOOST, MBMDR, Screen and Clean, SNPHarvester, SNPRuler, and TEAM were processed with many different settings of data sets. The results showed that BEAM3.0, BOOST, SNPHarvester, SNPRuler, and TEAM were the best algorithms in relation to Power, Type I Error Rate, and Scalability.The second stage is the creation of new combination of a new algorithm based on the results obtained in the first stage. This new algorithm is comprised of an aggregation of previously tested methodologies, of which 5 algorithms were chosen. This new Ensemble approach manages to maintain the Power of the best algorithm, while decreasing the Type I Error Rate.The purpose of this dissertation is to show a new complex empirical study, and a newly improved Ensemble approach, that shows better results than the existing state-of-the-art algorithms.