Benchmarking RNA velocity methods across 17 independent studies

Abstract

RNA velocity techniques offer great potential for unveiling trajectories of cell state transitions in different biological contexts. While diverse computational methods have been developed, there is no evidence-based guidelines for best-practice in RNA velocity inference. Here we conducted a benchmark study for 14 existing RNA velocity methods in 17 independent datasets. Many validations were done for the first time. We evaluated the performances of each method by measuring accuracy, stability, and usability. Our data showed no single method exhibited superior performance in all the assessments, and unexpected underperformance was observed in certain cases. Especially, the lack of uniformity in the inference results highlights the necessity to compare and control of multiple methods in a single analysis. Our study revealed current limitations and challenges in the RNA velocity methods and informed the best-practice for future studies.