Refining cell subpopulation identification and cell fate mapping using cell energy

Abstract

Despite rapid advances in single-cell data analysis algorithms, the accurate and robust mapping of cell fate dynamics across diverse biological systems remains a challenge. Here, we introduce CellEnergy, a novel method capable of assigning a cell energy indicator based on Gene Local Network Energy (GLNE), suitable for quantitatively describing attractor states. Benchmarking against 8 other advanced models, CellEnergy demonstrates superior performance in quantifying cellular plasticity, assessing cell population heterogeneity, automatically detecting terminal cell states, and tracking the dynamics of cell fate transitions. Moreover, CellEnergy shows robustness in rare cells and high dropout ratio datasets, all while significantly reducing computational costs. CellEnergy quantitatively evaluates the heterogeneity of mouse embryonic stem cells and identifies a cell subpopulation that exhibits consistent high differentiation potential, defining 2-cell-like cells. For mouse pancreatic development, CellEnergy accurately identifies four terminal cell states, including the rare Delta cells, and profiles the local network energy change trends of relevant genes across different lineages. Applications of CellEnergy reveal the impact of small molecules on cell plasticity from an energy perspective in human chemical reprogramming data. We further demonstrate the utility of CellEnergy in identifying plasticity-associated genes implicated in the tumorigenesis and prognosis of head and neck squamous cell carcinoma and melanoma. CellEnergy characterizes key branching points in human hematopoietic lineage commitment and decodes the distinct regulatory dynamics of erythropoiesis and megakaryopoiesis.