Characterizing transition cells in developmental processes from scRNA-seq data

Multi-cellular organism development involves orchestrated gene regulations of different cell types and cell states. Single-cell RNA-Seq, enable simultaneous observation of cells in various states, making it possible to study the underlying molecular mechanisms. However, most of the analytical methods do not make full use of the dynamics captured. Here, we model single-cell RNA-seq data obtained from a developmental process as a function of gene regulatory network using stochastic differential equations (SDEs). Based on dynamical systems theory, we showed that pair-wise gene expression correlation coefficients can accurately infer cell state transitions and validated it using mouse muscle cell regeneration scRNA-seq data. We then applied our analytical framework to the PDAC (Pancreatic ductal adenocarcinoma) mouse model scRNA-seq data. Through transition cells found in the pancreatic preinvasive lesions scRNA-seq data, we can better explain the heterogeneity and predict distinct cell fate even at early tumorigenesis stage. This suggests that the biomarkers identified by transition cells can be potentially used for diagnosis, prognosis and therapeutics of diseases.