Tumor subtype identification with weighted sparse non-negative matrix factorization for multiple heterogeneous data integration

Tumor subtype identification is an important topic for personalized medicine to design better treatment for each subcategory. It has been shown that non-negative matrix factorization (NMF) performs well for many practical problems including tumor subtype identification. However, input genes can also affect its performance. In this paper, we review a variation of sparse NMF (sNMF), and introduce a novel algorithm of the weighted sparse NMF (wsNMF) to incorporate known biological knowledge by integrating multiple heterogeneous data (e.g., gene expression, mutations, protein-protein interaction network, and transcription factor target network). wsNMF is applied to the identification of tumor subtypes of uterine corpus endometrial carcinoma.