Abstract
Recent advances in single-cell RNA sequencing technology provided unprecedented opportunities to simultaneously measure the gene expression profile and the transcriptional velocity of individual cells, enabling us to sample gene regulatory network dynamics along developmental trajectories. However, traditional methods have been challenged in offering a fundamental and quantitative explanation of the dynamics as differential equations due to the high dimensionality, sparsity, and complex gene interactions. Here, we present scDVF, a neural-network-based ordinary differential equation that can learn to model single-cell transcriptome dynamics and describe gene expression changes across time at a single-cell resolution. We applied scDVF on multiple published datasets from different technical platforms and demonstrate its utility to 1) formulate transcriptome dynamics of different timescales; 2) measure the instability of individual cell states; and 3) identify developmental driver genes upstream of the signaling cascade. Benchmarking with state-of-the-art vector-field learning methods shows that scDVF can improve representation accuracy by at least 50%. Further, our perturbation studies revealed that single-cell dynamical systems may exhibit properties similar to chaotic systems. In summary, scDVF allows for the data-driven discovery of differential equations that delineate single-cell transcriptome dynamics.
Teaser Using neural networks to derive the ordinary differential equations behind single-cell transcriptome dynamics.
Competing Interest Statement
The authors have declared no competing interest.