Systematically Identifying Genes and Pathways in Multiple Cancer Types Using HGD #x00026; PSO-SVM

摘要

Identification of genes and pathways which are risk factors for multiple cancers could help us prevent or treat cancer more effectively. Machine learning techniques have been extensively used to analyze microarray data but, most methods are based on the identification of significant associations of gene ontology terms with groups of genes. This does not directly reflect metabolic networks. In this paper, a more systematic approach is considered. As a first step, we did pathway analysis using Hyper Geometric Distribution (HGD) and significantly overrepresented sets of reactions (pathways or sub-pathways) were identified. As a second step, feature selection based Particle Swarm Optimization (PSO) and the K-Nearest Neighbor (K-NN) methods were used. We also used the Leave-One-Out Cross-Validation (LOOCV) as an evaluator of PSO and One-Versus-Rest (OVR) method as a component classifier. Experimental results show that our method simplifies features effectively and obtains higher classification accuracy than the other classification methods from the literature.