Clustering of Single Cells Based on Expressions of Subsets of Genes

• Activity Number: 123 - Unraveling Tissue Heterogeneity for Analyzing Omics Data in Cancer Research
• Type: Topic Contributed
• Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
• Sponsor: WNAR
• Author(s): Sha Cao* and Changlin Wan and Chi Zhang and Anru Zhang
• Companies: Indiana University School of Medicine and Purdue University and Indiana University School of Medicine and University of Wisconsin-Madison
• Keywords: Single cell RNA-Seq; Subspace clustering; Feature selection; Local low rank

Abstract

Single cell gene expression profiles exhibit high dimensional features of the cells, part of which contribute to cell type specificities and identities, while the rest are purely stochastic. Further, the count based single cell expression data are often subject to high drop-out events. These remain as the two challenges in clustering single cells and predicting their cell (sub)type identities. In other words, high dimensional gene features contain far too many noisy ones that tend to obscure the true cluster identities; and the drop out events make it challenging to directly apply clustering on the original data as their distance measures are highly sensitive to outliers. We assumed a composite model for single cell expression data based on multinomial and Poisson distribution. Then we turn to look for a sparse local low rank structure on the parameter space of the resulted entry-wise Poisson distributed matrix . This resulted in subspace clusters on the parameters space instead of on the original count data, which particularly handles the aforementioned two challenges in single cell clustering.