Predictive Modeling of Single Cell Transcriptomic Dynamics

Single-cell RNA sequencing (scRNA-seq) provides a powerful means to investigate cell fate dynamics during development and disease. However, existing computational approaches—such as RNA velocity—are often limited to qualitative and descriptive insights. We previously presented Dynamo, a machine learning framework that leverages noisy RNA velocity estimates to learn analytical and predictive vector fields of gene expression dynamics. These learned vector fields enable a suite of novel, quantitative analyses, including mapping the geometry of cell state transitions, identifying dynamic features of differentiation trajectories, inferring gene regulatory networks, predicting optimal paths and transcription factor cocktails for cellular reprogramming, and simulating in silico genetic perturbations. Here, we provide a comprehensive protocol for applying Dynamo across four distinct data types and four downstream analyses, demonstrating its versatility in handling conventional and metabolic labeling-enabled scRNA-seq, multi-omic datasets, and even datasets without velocity information. We also show that Dynamo can predict cell fate diversions following genetic perturbations and accurately identify key transcription factors that enable lineage transitions in the hematopoietic system. By enabling predictive modeling from single-cell data, our protocol aims to empower the development of generative and predictive models of single-cell genomics