Bursty RNA Velocity of Gene Programs for Trajectory Inference

Advances in single-cell RNA sequencing (scRNA-seq) allow collections of tens of thousands cells undergoing differentiation or experiencing external stimuli, revealing a detailed picture of stochasticity of gene expression. This permits trajectory inference (TI) which leverages variation in transcriptional profiles to assign an order of progression to cells, enabling a temporal interpretation of cell state. Traditional TI methods are data-driven and ignore the underlying biological processes so the directionality of developmental courses must be provided as inputs. While the standard RNA velocity method has provided additional insights, its moments-based coarse-grained setup has led to results that are contradictory to experiments. Here, we introduce a new RNA velocity model which accounts for transcriptional bursting and infers kinetic parameters using the joint distributions of counts. Furthermore, our model incorporates topic modeling, a popular method to explore gene programs, which helps delineate dynamical patterns of local gene programs not observed on the global scale. By disentangling the intertwining dynamics of genes, our method provides crucial mechanistic insights for cell fate commitments.