GSE

Abstract

RNA-seq contains rich information about individual even single cell, implies certain biology pattern vary in special time or space two dimensions, e.g. different life stage or environment. Byusing clustering and other computing methods, we can efficient analysis and decode those data applying to cancer diagnosis and treat, biological evolution and so on. However, RNA-seq data has features of super-high dimensions, less labeled samples and strong noise, which bring large challenges for clustering analysis. Therefore, we proposed a new clustering method GSE, which can efficient enhance the signal-to-noise ratio of input similarity matrix using diffusion process in weighted connection network to improve clustering performance. Comparing with latest clustering methods, our method has advantages in external clustering criterions NMI and ARI indicators. Meanwhile inadequacy and improved idea are given. Code can be downloaded from Git-hub.