Statistical inference with a manifold-constrained RNA velocity model uncovers cell cycle speed modulations

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Cells exhibit coordinated gene expression changes during biological processes, forming low-dimensional transcriptome dynamics.
• Single-cell RNA sequencing (scRNA-seq) captures temporal snapshots, while RNA velocity estimates gene expression dynamics using spliced and unspliced RNA.
• Existing RNA velocity algorithms lack statistical rigor and dynamic consistency with gene expression manifolds.

Approach:

• Developed a generative model for RNA velocity coupled with a Bayesian inference approach.
• Created VeloCycle, a unified framework for estimating velocity fields and gene expression manifolds.
• Implemented VeloCycle for studying cell cycle dynamics on periodic manifolds and validated with live imaging.

Key Points:

• VeloCycle provides statistically robust RNA velocity inference, overcoming limitations of previous methods.
• The model coherently identifies parameters of autonomous dynamical systems governing gene expression.
• Sensitivity analyses, one- and multiple-sample testing, and Markov chain Monte Carlo inference were performed for benchmarking.

Conclusions:

• VeloCycle accurately infers cell cycle periods and reveals gene-specific kinetics.
• Applied to in vivo and in vitro data, VeloCycle identified proliferation modes in neural progenitors and gene knockdown effects on cell cycle speed.
• VeloCycle enhances the scRNA-seq analysis toolkit with a modular and statistically rigorous framework for RNA velocity inference.