Topological-based classification using artificial gene networks

The topological properties of artificial gene networks that rely on the assessment of similarities between the expressions profile of gene pairs would be consistent with specific cellular states. Furthermore, some topological properties of the network would have significant variance regarding different cellular states. In this study, we proposed a novel and highly efficient computational framework for a topological-based classification using microarray gene expression data. Owing to the high prediction accuracy with this approach, we believe that there is a noticeable advantage of using topological-based classification. We used the microarray gene expression data sets containing 14 classes of cancer to construct 14 basic artificial gene networks accordingly. For each test sample, we add the sample into the data set of each class and reconstruct all of the networks. Cancer type was classified according to the correlation of topological quantity between the basic artificial gene networks and the reconstructed networks. Total classification accuracy can achieve 78.26% in the test data set (95.83% in the standard data set). After screening the quality of the standard data set, we can achieve a prediction accuracy of 86.48% in the test data set. Thus, we can achieve higher prediction accuracy in comparison to previous studies.