Spatially-Coupled Network RNA Velocities: A Control-Theoretic Perspective

Boya Hou, Maxim Raginsky, Abhishek Pandey, Olgica Milenković

This paper, recently uploaded to arXiv, presents a sophisticated mathematical bridge between the world of single-cell biology and the rigorous field of control theory. It introduces a new way to model RNA velocity—the "speed and direction" of gene expression—by accounting for the complex wiring both inside and between cells.

TL;DR

The authors propose a "spatially-coupled" RNA velocity model that explicitly incorporates gene regulatory networks (GRNs) and cell-to-cell communication. By applying control and consensus theory, they provide a framework to prove system stability and design optimal, time-efficient drug intervention strategies to steer cells toward desired states.

What Problem Does This Paper Address?

Standard RNA velocity models (like velocyto or scVelo) are powerful tools for predicting a cell's future state based on the ratio of unspliced to spliced mRNA. However, these models typically treat genes and cells as isolated units. They ignore two fundamental biological realities:

  1. Intracellular Wiring: Genes don't act alone; they activate or repress each other in a Gene Regulatory Network (GRN).
  2. Intercellular Communication: Cells "talk" to their neighbors via signaling molecules, a factor increasingly visible thanks to new spatial transcriptomics technologies.

Without accounting for these connections, current models struggle to predict how a localized perturbation (like a drug targeting one gene) will ripple through the entire system.

Methods

The authors move beyond simple linear models to a two-level networked ODE (Ordinary Differential Equation) system:

  • The GRN Layer: They use a rational "gene control function" (inspired by Hill functions) to model how the spliced RNA of "regulator" genes determines the transcription rate of "target" genes.
  • The Spatial Layer: They introduce a consensus term using a graph Laplacian. This models how cells exchange signaling molecules to synchronize their behavior, effectively treating the tissue as a multi-agent system.
  • Control Theory Analysis: The paper uses Lyapunov direct methods to derive strict conditions under which these biological systems reach a stable equilibrium.
  • Optimal Control: To model drug intervention, they frame it as a minimum-time optimal control problem. They use the Pontryagin Maximum Principle (PMP) to derive the optimal "dosage" strategy and use Lie brackets to determine if a specific target state is mathematically "reachable" given the network's topology.

Datasets

This paper is primarily a theoretical and computational study. As such, it does not introduce a new biological dataset. Instead, it uses:

  • Simulated GRN Topologies: The authors validate their model using synthetic 3-gene and 5-gene networks with known activation/repression links.
  • Synthetic Spatial Networks: They simulate a 5-cell population arranged in a complete graph to test how spatial coupling affects the spread of drug effects.

Key Insights and Ideas

  • Stability is a Balancing Act: The authors prove that for a stable equilibrium to exist, the "positive regulation" (activation) in the GRN cannot overwhelm the natural degradation rates of the RNA.
  • The Power of Expander Graphs: A major insight is that if the cellular communication network has "expander-like" properties (sparse but highly connected), the entire population can reach a functional consensus very quickly, even if only a few cells are directly perturbed.
  • "Bang-Bang" Interventions: The optimal control analysis reveals that the most efficient way to reach a target state is often a "bang-bang" strategy—switching abruptly between the maximum and minimum possible intervention levels rather than a gradual adjustment.
  • Bystander Effects: In spatially coupled models, the authors show that cells not targeted by a drug can still be "pulled" toward the target state by their neighbors, a mathematical representation of the bystander effect in therapy.

Key Results

  • Equilibrium Conditions: The paper provides a specific threshold (based on the spectral radius of a regulatory matrix) that guarantees a unique, globally stable equilibrium.
  • Consensus Bounds: They prove that the deviation of any single cell from the population average is upper-bounded by a value inversely proportional to the "algebraic connectivity" (the second smallest eigenvalue of the graph Laplacian) of the cellular network.
  • Simulation Success: In a 5-cell simulation where only 3 cells were treated with a "drug," the model successfully predicted how the 2 untreated cells would still show a decrease in target gene levels due to intercellular interconnection.

Limitations and Future Work

  • Parameter Inference: The current model assumes we already know the "weights" of the GRN and the communication rates. In the real world, these must be learned from noisy, high-dimensional sequencing data, which is a massive computational challenge.
  • Deterministic vs. Stochastic: The authors use deterministic ODEs. However, gene expression is inherently "noisy" (stochastic). Future work could extend this to Stochastic Differential Equations (SDEs).
  • Scaling: While the math is rigorous, scaling these control-theoretic approaches to thousands of genes and millions of cells will require significant algorithmic optimization.

Reference: Hou, B., Raginsky, M., Pandey, A., & Milenkovic, O. (2026). Spatially-Coupled Network RNA Velocities: A Control-Theoretic Perspective. arXiv:2601.01018v1.

Summary updated

Abstract

RNA velocity is an important model that combines cellular spliced and unspliced RNA counts to infer dynamical properties of various regulatory functions. Despite its wide applicability and many variants used in practice, the model has not been adequately designed to directly account for both intracellular gene regulatory network interactions and spatial intercellular communications. Here, we propose a new RNA velocity approach that jointly and directly captures two new network structures: an intracellular gene regulatory network (GRN) and an intercellular interaction network that captures interactions between (neighboring) cells, with relevance to spatial transcriptomics. We theoretically analyze this two-level network system through the lens of control and consensus theory. In particular, we investigate network equilibria, stability, cellular network consensus, and optimal control approaches for targeted drug intervention.
BibTeX
@misc{hou2026spatially,
  author = {Boya Hou and Maxim Raginsky and Abhishek Pandey and Olgica Milenković},
  title = {{Spatially-Coupled Network RNA Velocities: A Control-Theoretic Perspective}},
  year = {2026},
  eprint = {2601.01018},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.01018},
}
Figure