Learning feedback loops in transcriptome network during cell fate transition

Abstract

Embryonic development is characterized by series of cell fate transitions, during which non-linear feedback of signaling events and network dynamics in the transcriptome are instrumental to determine differentiation outcomes. Changes in selected genes or pathways have been described extensively, but a system-level understanding of gene regulatory network regulation is still lacking. Leveraging time-series data from single-cell transcriptome profiling of developing mouse embryos, we collapsed dynamic gene expressions into changes in 14 core biological processes, and constructed an ordinary differential equation-based machine learning model of the transcriptomic network to capture its feedback structure. We evaluated the polarity and magnitude of direct pairwise causal relationships between biological processes and identified higher-order feedback loops that dominate system behavior. Despite heterogeneous expressions at the gene level, we find that the transcriptome at the macroscale level has a feedback structure that is intrinsically stable and robust. We show the pivotal role of intracellular signaling in driving systemic changes and uncover the importance of homeostatic process and establishment of localization in regulating network dynamics. Local regulatory structures represent domain-specific regulations that are essential to cell fate transition, especially lipid metabolic process. Together, this study provides a holistic picture of transcriptomic network dynamics during mouse organogenesis, offering insight into key aspects of information flow in the transcriptome that control cell fate transition. Conflict of interest statementThe authors declare no potential conflicts of interest. FundingThe authors received no funding for this study. Author contributionsConceptualization: LX; Methodology: LX, WAS, and JDS; Data Curation: LX; Investigation: LX and WAS; Formal Analysis: LX and WAS; Visualization: WAS and LX; Writing - review & editing: LX and WAS.