Heuristic Breadth-First Search Algorithm for Informative Gene Selection Based on Gene Expression Profiles

Abstract

The tumor diagnosis method based on gene expression profiles will be developed into the fast and effective method in clinical domain in the near future. Although DNA microarray experiments provide us with huge amount of gene expression data, only a few of genes are related to tumor in gene expression profiles. Moreover, it is difficult to select informative genes related to tumor from gene expression profiles because of its characteristics such as high dimensionality, small sample set and many noises in gene expression profiles. According to its characteristic, a novel heuristic breadth-first search algorithm based on support vector machines is proposed, which can simultaneously find as many informative gene subsets as possible in which the number of informative genes is almost least but its classification performance is almost highest in spite of its time-consuming characteristic. Three tumor sample sets are examined by the novel approach and experiments show that the novel approach is feasible and effective in tumor classification. Experiment results show that 100% of 4-fold cross-validation accuracy has been achieved by only two, four and four genes for leukemia, colon tumor and SRBCT (Small Round Blue Cells Tumor) datasets, respectively, which is superior to the results of other tumor classification methods. To avoid the affect of different partition of sample set, the full-fold cross-validated method that can more objectively evaluate the classification performance of informative gene subset is proposed.