What is it about?

Understanding how living cells develop requires knowing which genes control others, i.e. the gene regulatory network (GRN). Currently, scientists use single-cell sequencing to study this, but the data comes in static snapshots, like a series of photos of crowds walking on a pedestrian, rather than a continuous video. Furthermore, this data is often messy, with many missing values (a problem known as "dropout"). In this study, we developed FlowGRN, a new computational tool that uses advanced AI tool called flow matching to solve these problems. Think of it as using AI to fill in the missing frames between photos to create a smooth movie of cell development. By reconstructing these cellular trajectories and taking consideration of the data noise, FlowGRN allows us to reconstruct the interactions between thousands of genes from the recovered cell trajectories. This provides a clearer picture of how cells make decisions during development and disease.

Featured Image

Why is it important?

Flow matching is originally a AI tool to model distributional shift in generative AI, such as generating images from gaussian noise. We noticed that the same tool is applicable on gene expression data and cell dynamic modelling. Under this framework, we can reconstruct the continuous developmental paths of cells, effectively filling in the missing timeline between static data snapshots. By analyzing the mathematical "flow" that drives these cellular transitions and existing dynGENIE3 method, we can deduce the direction and strength of gene interactions, which tell us not just which genes are active, but which ones are actually controlling the process. This makes it possible to uncover comprehensive regulatory networks in complex biological systems, which is essential for understanding genetic disorders and identifying potential drug targets.

Perspectives

Gene interaction networks act as a high-level summary of the complex biological mechanisms operating within our bodies. Consequently, formalizing and constructing models to accurately simulate these interactions has been a long-standing scientific challenge. Flow matching provides us with a uniquely flexible framework to model these dynamics, and this paper represents a successful trial of applying this concept to single-cell data. I am very excited to dig deeper into this direction, using these AI-driven tools to decipher the precise rules governing cell dynamics.

Tsz Pan Tong
Universite du Luxembourg

Read the Original

This page is a summary of: FlowGRN: Scalable and Dropout-Robust Gene Regulatory Network Inference via Flow Matching-Based Trajectory Reconstruction, October 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3765612.3767196.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page