Presentation Information

[SS16-02]RNA landscape in single cells

*Yusuke Imoto1 (1. Kyoto University (Japan))

Keywords:

scRNA-seq,RNA velocity,Landscape,Hodge decomposition,Trajectory

Understanding the complex dynamics of cell differentiation is a key challenge in biology. Cells differentiate into different cell types by controlling the expression levels of numerous genes. That is, cell differentiation can be regarded as high-dimensional dynamics formed by genes. Dr. C. H. Waddington suggested simplifying such complex dynamics as a terrain with multiple-branching paths, called "landscape." Waddington's landscape visualizes the existence of bifurcations and when and where bifurcations occur, leading to our understanding of the complex dynamics. In this talk, we introduce a novel framework for constructing a landscape from single-cell RNA-seq data based on a Hodge decomposition. Single-cell RNA-seq data quantify RNA expression levels of whole genes using a next-generation sequencer in single-cell resolution, being high-dimensional matrix data because the number of genes is approximately 20,000 in humans. Additionally, single-cell RNA-seq data can be transformed to velocity data using the RNA velocity technique. Our framework extracts the potential of gradient flow by decomposing the velocity data using a Hodge decomposition and constructs the landscape based on that potential. The height of the landscape adopts the potential as it is, the vertical component uses pseudo-time along the potential decay, and the horizontal component employs the most variable component in the original gene expression space. This landscape design makes us intuitively understand when, where, and how bifurcations occur. We applied this framework to single-cell gene expression data from primordial germ cell-like cell (PGCLC) induction, a process that models early germ cell development. The landscape uncovered two major bifurcations in pseudo-time, identifying key regulatory genes involved in cell fate decisions. Differential gene expression analysis of bifurcation regions provided further biological validation.