用于样本聚类和网络分析的整合鲁棒结构化NMF模型

In order to preserve the homogeneity among data more effectively, this paper proposes an integrated robust structured non-negative matrix factorization (integrated robust structured non-negative matrix factorization, iRSNMF) model with an induced structured term. We verify the effectiveness of this model by applying it to the clustering experiments of cancer samples and the analysis of gene co-expression network. Reasonable biological explanations of related genes and pathways are given based on existing literature. Experimental results show that the iRSNMF method has excellent clustering performance and more-key genes mining ability. The genes and pathways obtained by the iRSNMF model play an important role in cancer pathogenesis, accordingly, providing a new idea for the diagnosis, treatment and prognosis of cancer.