A relay velocity model infers cell-dependent RNA velocity

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables the study of cellular dynamics.
• RNA velocity aims to predict cell fate by analyzing spliced and unspliced mRNA levels.
• Existing RNA velocity models struggle with complex cellular transitions due to universal kinetic assumptions.

Purpose of the Study:

• To develop a scalable deep neural network, cellDancer, for accurate, single-cell resolution RNA velocity inference.
• To overcome limitations of conventional models in multi-stage and multi-lineage cell differentiation.
• To identify cell-specific kinetic rates as potential cell fate indicators.

Main Methods:

• cellDancer employs a deep neural network to infer local RNA velocity from neighboring cells.
• The model aggregates local velocities for a global, single-cell resolution kinetic inference.
• Performance was benchmarked using simulations with varying kinetic regimes, dropout rates, and data sparsity.

Main Results:

• cellDancer demonstrated robust performance across diverse simulation scenarios, including sparse and high-dropout datasets.
• The model successfully modeled complex biological processes such as erythroid maturation and hippocampus development.
• cellDancer provided cell-specific predictions of transcription, splicing, and degradation rates.

Conclusions:

• cellDancer offers a scalable and accurate approach for RNA velocity analysis in scRNA-seq data.
• The method effectively handles complex cellular dynamics where traditional models fail.
• Cell-specific kinetic rates predicted by cellDancer may serve as valuable biomarkers for cell fate determination.