Abstract:
|
Each tumor is composed of multiple cell types, each one characterized by a particular genomic and transcriptomic profile. With the advancement of single cell sequencing technology, gene expression data can now be measured at single cell resolution. Based on cell-specific transcriptomic profiles, cells can be classified into different types including tumor, immune and stromal cells. This classification is utilized to quantify the proportion of each cell type in tumor tissue and fully capture the tumor heterogeneity across different patients. Here, we propose a repulsive Bayesian mixture model which can efficiently classify cells into different subtypes and estimate the proportion of each cell type for different patients. Specifically, each cell type is captured by a mixture component with the identifiability of different mixture components being induced by the repulsive nature of the prior. In addition, the proposed Bayesian framework can naturally deal with the sparseness of the data. The performance of the proposed approach is evaluated on various synthetic data examples as well as single cell sequencing data for different cancer types.
|