Combine Pathway Analysis with Random Forests to Hunting for Feature Genes

Abstract

In this paper, a method combining pathway analysis with random forests was provided. After the important pathways were discovered by computing the classification error rates of out-of-bag (OOB), the feature genes were also discovered according to these important pathways. The important pathways were recombined as the new gene sets and the classification error rates were recomputed by random forests algorithms. According to the rank and the frequency of feature genes, the important feature genes associated with disease were discovered. At each important pathway, the relativity of gene expression was also studied. The results showed that our method was available because the expressions of genes at the same pathway were approximate. Those genes selected by SAM software directly were not feature genes but noises. We also compared random forests with other machine learning methods and found that random forests classification error rates were the lowest. This method can provide biological insight into the study of microarray data. Keywordsrandom forest, KEGG, pathway analysis, Microarray