Microarray data classification based on computational verb

Abstract

Computational verb (CV) theory is a relatively new research field in mathematics and has been applied to many different fields. In the field of pattern recognition, the CV-based rule induction algorithm can generate some simple rules with CVs and adverbs by linguistically interpretable forms. In this paper, we present an interpretable rule extraction framework based on CV rule theory for the classification of microarray data. In contrast to the existing rule-based methods, the CV method enables to explicitly express the relationships of the genes based on some mathematical templates and hence enhance the understanding on the data results. Stay is a typical verb used in the CV to describe the trend of changes. In our algorithm, Stay is applied to generate CVR by a gene pair, named SCVR. The corresponding evolving and similarity functions for calculating the difference between SCVR rules are also presented to illustrate this process. Similar to other rule-based methods, the SCVR can achieve significant gene selection and cancer classification task concurrently. To evaluate the performance of our proposed approach, we conduct the experiments on several binary class and multiclass microarray datasets. Experiments confirm that the proposed method can outperform many rule-based classiers with the fusion of five rules.