The study of dynamic biological phenomena, such as differentiation, immune responses, and cancer - which involve multiple simultaneously occurring biological processes - using destructively measured transcriptomic profiles remains a central challenge in single-cell data science. RNA velocity estimation is a potentially powerful tool to reveal the directionality of transcriptional changes in single-cell RNA-seq data, but it lacks accuracy, absent advanced metabolic labeling techniques. To empower RNA velocity for more general systems, including immune responses, we developed a new approach that reflects our finding that parameter estimation depends critically on the input cells and genes. While these are typically globally selected, we use probabilistic topic models, a type of latent factorization, to infer the cells and genes specifically relevant for estimating process-specific velocity parameters. We are now developing new methods that account for the stochastic nature of lineage commitment, and we are investigating how to use RNA velocity to gain mechanistic insights into cell fate decisions.

Who is involved: Frank Gao