Quantifying uncertainty in RNA velocity.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables dynamic analysis via RNA velocity.
• Existing RNA velocity methods often lack uncertainty quantification and rely on complex, uninterpretable models.
• Unrealistic assumptions in current models limit their biological applicability.

Purpose of the Study:

• To develop a Bayesian hierarchical model for RNA velocity estimation with improved interpretability and uncertainty quantification.
• To address limitations of existing methods, including unrealistic assumptions and lack of uncertainty estimation.
• To provide a robust framework for inferring dynamic information from scRNA-seq data.

Main Methods:

• A Bayesian hierarchical model incorporating time-dependent transcription rates and non-trivial initial conditions.
• Discussion of model parameter identifiability, including latent time.
• A novel algorithm combining Markov chain Monte Carlo and consensus approaches for full Bayesian inference and uncertainty quantification.

Main Results:

• The proposed Bayesian model provides well-calibrated uncertainty quantification for RNA velocity estimates.
• Identifiability of model parameters, including larger latent time values, is addressed.
• Validation through comprehensive simulations and comparison with existing RNA velocity methods on mouse embryonic stem cell data.

Conclusions:

• The novel Bayesian approach offers reliable RNA velocity estimation with robust uncertainty quantification.
• The method's interpretability and ability to handle complex biological scenarios are demonstrated.
• Results align with cell cycle phases, highlighting the model's biological relevance for dynamic single-cell analysis.