FlowGRN: reconstructing the dynamics of gene regulation

How do genes control one another as a cell develops, differentiates, or transforms into disease? Understanding this complex interplay — known as gene regulatory networks (GRNs) — is key to explaining how stem cells become neurons, or how cancer cells develop themselves. 

Until now, scientists have been studying how genes interact using single-cell RNA sequencing (scRNA-seq), which lets them see the activity of thousands of individual cells. But this method only captures snapshots — like photos of cells at different stages. Without seeing the full journey, it’s hard to figure out which genes are really driving the process. 

Understanding gene changes withFlowGRN

To address this challenge, doctoral student Tsz Pan Tong and his supervisor Prof. Jun Pang have developed FlowGRN, a novel computational method that leverages recent advances in conditional flow matching (CFM), a modern machine learning technique that helps models learn realistic transitions between different states. In this context, it enables the reconstruction of how cells evolve from one state to another. FlowGRN uses CFM to learn these hidden cellular dynamics, reconstruct cell trajectories from scRNA-seq data, and infer how genes influence one another along these developmental paths. 

In extensive tests on both simulated and experimental datasets, FlowGRN achieved high precision and ranked among the top performers in the BEELINE benchmark — a widely used framework that objectively compares gene network inference methods. The method also includes strategies to mitigate the impact of measurement errors in the data, helping it focus on the true biological signals. Importantly, FlowGRN can distinguish between genes that activate others (turning them “on”) and those that suppress them (keeping them “off”), offering a more complete picture of how cellular processes are regulated. 

Medical applications

By understanding how cells transform over time, FlowGRN helps scientists better understand the genetic changes that drive development and disease. These insights can make experiments more focused and efficient, and in the long run, support advances in precision medicine. For example, it could point to promising drug targets or help anticipate how patients might respond to treatment. 

“Biologists ultimately want to see what a cell is doing,” explains Tsz Pan. “If we can capture how stem cells grow into muscle or neural cells and understand what goes wrong when they become cancerous, we can start to decode the logic of life at the cellular level and find ways to prevent these transformations.” 

Recently, the project drew international attention at the ACM BCB 2025 conference in Philadelphia, where the team’s paper earned a Best Paper Honorable Mention Award. This recognition underlines the quality and potential impact of their work. “This award strengthens our confidence in refining FlowGRN,” comments doctoral student Tsz Pan. “It’s also an important step toward a more data-driven understanding of how cells work.”, adds Prof. Jun Pang.  

Looking ahead, the researchers aim to broaden FlowGRN’s scope by applying it to other types of single-cell data, such as chromatin accessibility, which shows how DNA is folded and which genes are open for use, and spatial omics, which reveals where gene activity occurs inside tissues. 

The project is funded under the Audacity Grant (AUDACITY-2021) from the Institute for Advanced Studies (IAS) and under INTER/NCN/24/18732364/EdgeCR from the Fonds National de Recherche (FNR).