Dissecting and steering cell dynamics using spatially-informed RNA velocity with veloAgent

Area of Science:

• Single-cell transcriptomics
• Systems biology
• Computational biology

Background:

• RNA velocity infers cell state transitions by modeling mRNA dynamics.
• Current methods lack spatial context and scalability for large datasets.
• Tissue organization and dynamic processes remain challenging to study.

Purpose of the Study:

• Introduce veloAgent, a deep generative and agent-based framework.
• Integrate spatial information into RNA velocity analysis.
• Enable scalable and accurate inference of cell state transitions.

Main Methods:

• Agent-based simulations of local microenvironments to integrate spatial data.
• Deep generative modeling of transcriptional kinetics.
• In silico perturbation module for simulating regulatory interventions.

Main Results:

• veloAgent improves RNA velocity accuracy by integrating molecular and spatial cues.
• Achieves sublinear memory scaling for efficient analysis of large, multi-batch spatial datasets.
• Accurately estimates gene- and cell-specific transcriptional kinetics.

Conclusions:

• veloAgent is a scalable and versatile framework for dissecting spatially resolved cellular dynamics.
• Facilitates prediction of cell fate dynamics and regulatory intervention impacts.
• Advances the study of tissue organization and dynamic biological processes.