New AI framework predicts how cells make fate decisions

Scientists at the Stowers Institute for Medical Research, Helmholtz Munich, the Technical University of Munich, and the University of Oxford have developed RegVelo - a new AI framework that simultaneously models cellular dynamics and gene regulation, enabling researchers to predict, simulate, and experimentally validate how cells make fate decisions. The study is published in Cell on May 11, 2026.

How does a single cell know to become a pigment cell, a blood cell, or a neuron? Scientists have long been able to map where cells are headed. What has remained far harder is understanding which molecular regulators steer them there - and what happens when those regulators are altered.

Now, a team of researchers has developed RegVelo, an AI-based model that closes that gap. Published May 11, 2026, in Cell, the framework jointly learns how cells change over time and which gene regulatory networks drive those changes - making it possible to model both the trajectory and the molecular engine behind it.

Connecting two fields that have worked in parallel

Single-cell biology has given researchers increasingly detailed maps of development. RNA velocity methods estimate how cells move through developmental landscapes; gene regulatory network approaches identify the relationships among genes. Until now, these methods have largely been used in parallel rather than together.

RegVelo bridges that divide. Building on RNA velocity - which infers a cell's direction of change from the ratio of immature to processed RNA - the model adds a critical layer: It accounts for genes not as independent units but as parts of a network, where each gene can activate or suppress others. The result is a framework that can trace developmental trajectories and simulate the consequences of specific regulatory interventions.

For a long time, cellular dynamics and gene regulation have largely been modeled separately. RegVelo brings those pieces together, allowing us to ask not only how cells are changing, but which regulatory interactions are helping drive those changes."

Prof. Fabian J. Theis, co-senior author of the study and Director of the Computational Health Center (CHC) at Helmholtz Munich and Professor at the Technical University of Munich (TUM)

A collaboration built on complementary strengths

RegVelo emerged from a collaboration that integrated experimental and computational expertise across institutions. Tatjana Sauka-Spengler, Ph.D., co-senior author and Investigator at the Stowers Institute for Medical Research, moved her lab to Stowers from the University of Oxford in 2022. Sauka-Spengler, who remains a Visiting Professor at the MRC Weatherall Institute of Molecular Medicine, University of Oxford, contributed high-resolution gene regulatory circuitry from her team's research on cranial neural crest development. Theis's group at Helmholtz Munich brought computational tools for modeling single-cell trajectories and RNA velocity. First author Weixu Wang, a doctoral researcher at the CHC, led the development of the unified deep learning framework.

"What made this work especially powerful was the combination of complementary strengths," said Sauka-Spengler. "High-resolution gene regulatory circuitry from our lab, and dynamic trajectory and network modeling from Fabian's team, who are experts in what they do. RegVelo emerged from integrating those two views into one framework for the first time."

Predictions validated in zebrafish

The team tested RegVelo across multiple biological systems, including the cell cycle, blood cell formation, and pancreatic development. The most detailed case study focused on zebrafish neural crest cells - a versatile population of embryonic cells that give rise to pigment cells, nerve cells, and craniofacial tissues.

RegVelo identified tfec as an early driver of pigment cell development and revealed elf1 as a previously unknown regulator of pigment cell fate. Both predictions were validated experimentally through CRISPR/Cas9 knockout and single-cell Perturb-seq, demonstrating that the model can generate biologically meaningful hypotheses that hold up in living systems.

"Development is often described as a series of static snapshots of cell states," said Sauka-Spengler. "What we really want to understand is how cells make decisions - how they transition from one state to another. RegVelo models how those fate decisions are encoded in gene regulatory networks over time, and what drives them."

"RegVelo makes visible what consequences it has for a cell's developmental path when a specific genetic regulator is switched off," said Wang. "Verifiable predictions can be derived from single-cell data about which genetic regulators promote, slow down, or redirect a particular developmental path."

A step toward virtual cell models

The researchers describe RegVelo as a step toward a more predictive form of developmental biology, one in which computational models help prioritize experiments, uncover hidden regulators, and forecast how cell fates may shift when gene networks are perturbed. Looking further ahead, the approach could help researchers better understand disease-relevant cell states and identify new therapeutic targets - including in developmental disorders, cancer biology, and regenerative medicine.

"RegVelo is a step toward virtual cell models that will help us better understand how cells behave in differentiation contexts and how they respond to genetic perturbation," said Theis. "In the long term, this could help us identify possible starting points for new therapies."

"Having a full resolution of gene regulatory circuitry that has been predicted, simulated, perturbed, and validated gives us a very solid tool," Sauka-Spengler added. "We can start from stem cells or naïve cells and develop new ways of directing them toward cell types that can be used in cell therapies."