Multivariate stochastic modeling for transcriptional dynamics with cell-specific latent time using SDEvelo.

Area of Science:

• Computational Biology
• Genomics
• Single-cell analysis

Background:

• RNA velocity analysis in single-cell RNA sequencing (scRNA-seq) predicts cell differentiation trajectories.
• Existing dynamic modeling methods often use ordinary differential equations (ODE) for individual genes, neglecting multivariate and stochastic transcriptional dynamics.

Purpose of the Study:

• To introduce SDEvelo, a novel generative approach for inferring RNA velocity using multivariate stochastic differential equations (SDE).
• To address limitations of ODE-based models by explicitly accounting for transcriptional stochasticity and estimating cell-specific latent time.

Main Methods:

• Developed SDEvelo, a generative model employing multivariate stochastic differential equations (SDE) to analyze unspliced and spliced RNA dynamics.
• Integrated estimation of a gene-shared, cell-specific latent time to capture transcriptional progression.
• Validated SDEvelo on simulated data and diverse scRNA-seq and spatial transcriptomics datasets.

Main Results:

• SDEvelo accurately models stochastic transcriptional dynamics, including random patterns in mature cells.
• The method successfully detects carcinogenesis, demonstrating its utility in identifying disease states.
• The inferred gene-shared latent time aids downstream biological discovery and analysis.

Conclusions:

• SDEvelo offers a robust and scalable approach for RNA velocity inference, outperforming traditional ODE methods.
• The model's ability to handle transcriptional uncertainty and its applicability to both scRNA-seq and spatial transcriptomics enhance its utility in biological research.