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

Area of Science:

• Single-cell genomics
• Systems biology
• Computational biology

Background:

• Cells exhibit coordinated gene expression changes, forming transcriptome dynamics within low-dimensional manifolds.
• Current RNA velocity algorithms are often heuristic, lack statistical control, and produce dynamically inconsistent vector fields.
• Accurate inference of cellular dynamics is crucial for understanding biological processes.

Purpose of the Study:

• To develop a statistically consistent and robust Bayesian framework for RNA velocity estimation.
• To couple velocity field and manifold estimation into a unified model for dynamical system parameter identification.
• To provide an improved tool for analyzing single-cell RNA sequencing data, particularly for cell cycle and gene regulation studies.

Main Methods:

• Introduced a Bayesian model for RNA velocity that integrates velocity field and manifold estimation.
• Developed VeloCycle, a computational framework implementing the Bayesian model.
• Applied VeloCycle to study cell cycle dynamics on periodic manifolds and analyze gene regulation in response to knockdowns.

Main Results:

• VeloCycle provides a statistically consistent RNA velocity inference framework.
• The model successfully infers cell cycle periods using live imaging data.
• Demonstrated VeloCycle's utility in revealing speed differences in progenitors and gene knockdowns (Perturb-seq).

Conclusions:

• VeloCycle offers a modular and statistically sound approach to RNA velocity analysis.
• The framework enhances the toolkit for single-cell RNA sequencing data interpretation.
• Improved inference of cellular dynamics can advance our understanding of developmental processes and disease.