NeuroVelo: interpretable learning of temporal cellular dynamics from single-cell data

Reconstructing temporal cellular dynamics from static single-cell transcriptomics remains a major challenge. Methods based on RNA velocity are useful, but interpreting their results to learn new biology remains difficult, and their predictive power is limited. Here we propose NeuroVelo, a method that couples learning of an optimal linear projection with non-linear Neural Ordinary Differential Equations. Unlike current methods, it uses dynamical systems theory to model biological processes over time, hence NeuroVelo can identify what genes and mechanisms drive the temporal cellular dynamics. We benchmark NeuroVelo against several state-of-the-art methods using single-cell datasets, demonstrating that NeuroV-elo has high predictive power but is superior to competing methods in identifying the mechanisms that drive cellular dynamics over time. We also show how we can use this method to infer gene regulatory networks that drive cell fate directly from the data.