Deep Subspace Similarity Fusion for the Prediction of Cancer Subtypes

Prediction of cancer subtypes is an important basic task in cancer disease diagnosis and therapy. There have been a number of attempts to predict cancer clinical-type by using one type of data from molecular layers of biological system. Though some effects have been obtained, it still remains difficult to bridge the cancer genome to cancer phenotypes, because the genome is neither simple nor independent but rather complicated and dysregulated from multiple molecular mechanisms. Therefore many methods centered on the integration of diverse omics data to improve the understanding of tumorigenesis. The Similarity Network Fusion (SNF) is one of the most promising integrative clustering technique. However, SNF adopts Euclidean distance to measure the similarity between patients, which shows some limitations. In this paper, an improved version of SNF, namely Deep Subspace Similarity Fusion (DSSF), is proposed. DSSF utilizes auto-encoder and data self-expressiveness approaches to guide a deep subspace model, which can achieve effective expression of discriminative similarity between patients. As a result, the dissimilarity between inter-cluster is delivered and enhanced compactness of intra-cluster is achieved at the same time. The validity of DSSF is examined by extensive simulations over the subtypes prediction for five different cancer through three levels omics data. Clustering evaluations and survival analysis both demonstrate that DSSF delivers comparable or even better results than many state-of-the-art integrative methods.