A Dictionary Learning Algorithm for Gene Expression Profile Classification Based on Feature Selection

Abstract

The classification of gene expression profiles has become an important means of cancer classification. As a new machine learning method, dictionary learning has become more and more prevalent in the classification of gene expression profiles. In this paper, we propose a new dictionary learning framework based on feature selection. We first use training samples and their tag information to select those key gene sets that are helpful in classification. And we believe that these key gene sets are equally applicable to test samples. The process of dictionary learning is also based on these key gene sets. In dictionary learning, we train a sub-dictionary for each class of samples, and we also train a projection matrix P that can expand the distances of different classes of samples. Both of these operations can increase the final classification discriminability. The final experimental results show that the proposed method performs better classification on multiple data sets than the other dictionary learning methods or machine learning methods used in the experiments.