A Unified Multi-View Clustering Method Based on Non-Negative Matrix Factorization for Cancer Subtyping

Abstract

Non-negative matrix factorization (NMF) has gained sustaining attention due to its compact leaning ability. Cancer subtyping is important for cancer prognosis analysis and clinical precision treatment. Integrating multi-omics data for cancer subtyping is beneficial to uncover the characteristics of cancer at the system-level. A unified multi-view clustering method was developed via adaptive graph and sparsity regularized non-negative matrix factorization (multi-GSNMF) for cancer subtyping. The local geometrical structures of each omics data were incorporated into the procedures of common consensus matrix learning, and the sparsity constraints were used to reduce the effect of noise and outliers in bioinformatics datasets. The performances of multi-GSNMF were evaluated on ten cancer datasets. Compared with 10 state-of-the-art multi-view clustering algorithms, multi-GSNMF performed better by providing significantly different survival in 7 out of 10 cancer datasets, the highest among all the compared methods.