CLADES: a hybrid NeuralODE-Gillespie approach for unveiling clonal cell fate and differentiation dynamics

Recent lineage tracing based single-cell techniques (LT-scSeq), e.g., the Lineage And RNA RecoverY (LARRY) barcoding system, have enabled clonally resolved interpretation of differentiation trajectories. However, the heterogeneity of clone-specific kinetics remains understudied, both quantitatively and in terms of interpretability, thus limiting the power of barcoding systems to unravel how heterogeneous stem cell clones drive the overall cell population dynamics. Here, we present CLADES, a NeuralODE-based framework to faithfully estimate the clone and population-specific kinetics from both newly generated and publicly available LARRY LT-scSeq data. By incorporating a stochastic simulation algorithm (SSA) and differential expression gene (DEGs) analysis, CLADES yields the summary of cell division dynamics across differentiation time-courses and reconstructs the lineage tree of the progenitor cells in a quantitative way. Moreover, clone-level behaviors can be grouped into characteristic types by pooling individual clones into meta-clones for analyses at various resolutions. Finally, we show that meta-clone specific cellular behaviors identified by CLADES originate from hematopoietic stem and progenitor cells in distinct transcriptional states. In conclusion, we report a scalable approach to robustly quantify clone-specific differentiation kinetics of cellular populations for time-series systems with static barcoding designs.