A Novel Approach for Classifying Human Cancers

Abstract

Various researches have shown that machine learning approaches can be successfully used to detect and classify cancer tissue samples by their gene expression patterns. In this paper, an entropy-based improved k-TSP method (Ik-TSP) is proposed. We calculate the entropy for each gene based on the gene expression profile, and then find the best threshold of entropy depending on LOOCV accuracy for each gene expression dataset. Finally we select key genes for each gene expression dataset according to the best threshold and use them to implement Ik-TSP method to classify the cancer. Compared to 7 cancer classifiers mentioned in this paper in 9 binary public gene expression datasets of human cancers, the Ik-TSP method achieves an average LOOCV accuracy of 95.39%, and improves 3% better than the k-TSP method. Simulated experimental results show that the proposed Ik-TSP method is applicable to classify human cancers.