AI Tool Predicts Cell Fate, Revealing Hidden Development Drivers

RegVelo, an innovative AI technology created by scientists at the Stowers Institute and Helmholtz Munich, empowers researchers to not only predict the journey of cells as they acquire specific identities, but also to understand the pathways and drivers behind these transformations — offering crucial insights into developmental disorders, tumor progression, and regenerative medicine.

KANSAS CITY, Mo. and MUNICH, May 11, 2026 /PRNewswire/ — What determines the initial steps that guide a cell to transform into a blood cell, neuron, or pigment cell? While scientists have developed powerful methods to monitor these changes, the challenge remains to comprehend not just the direction cells are headed, but also the regulators that influence their ultimate fate.

Recent research from the Stowers Institute for Medical Research and Helmholtz Munich has introduced a revolutionary AI framework, RegVelo, published in Cell on May 11, 2026, that aims to clarify these questions. RegVelo merges two fields of single-cell biology that have previously operated separately: techniques for estimating cellular changes over time and methods for deciphering the gene regulatory networks that govern these transformations.

By integrating these components, RegVelo grants researchers the ability to ‘time travel’ through simulations, enabling predictions of cellular evolution and identification of the genes responsible for those changes, thus reducing the need for exhaustive laboratory experiments.

“Why is this significant?” inquired Tatjana Sauka-Spengler, Ph.D., a key investigator at the Stowers Institute and co-senior author of the study. “If we equip a very early set of cells with specific instructions, we could potentially recreate these cell types in vitro, closely resembling their natural development. These cells could then serve pivotal roles in cell therapies aimed at advancing regenerative medicine.”

Video: Predicting cell fate, hear from the scientists behind the work

“Dr. Sauka-Spengler and her team have pioneered a significantly distinct approach to data processing,” said Stowers President and Chief Scientific Officer Alejandro Sánchez Alvarado, Ph.D.. “This framework allows us to hypothesize the most probable trajectory of each component over time and space, enabling us to utilize deep learning for dynamic predictions and experimental validation.”

In their study, RegVelo was employed to model the neural crest, a group of embryonic cells capable of differentiating into various body structures. In the development of zebrafish neural crest, RegVelo pinpointed an early driver of pigment cell formation (tfec) and uncovered a novel regulator of pigment cell fate (elf1). These predictions were subsequently corroborated through experiments, validating the model’s capacity to do more than merely describe developmental transitions.

“There is always an initiating, driving element that shapes a final outcome,” stated Sauka-Spengler. “However, these pivotal elements are often overlooked when only analyzing the endpoint of cell development. Normally, development is depicted as a series of static cell state snapshots, but our focus should be on understanding how cells make decisions and transition from one state to another. RegVelo models the encoding of these fate decisions within gene regulatory networks across time and space, as well as the factors driving them.”

By linking early regulatory events to later cellular fates, this research may enhance the study of developmental disorders and progressively inform regenerative medicine and cell therapy strategies.

“The utility of RegVelo extends far beyond neural crest cells,” emphasized Sánchez Alvarado. “It can be applied to any system in which cells undergo change over time, ranging from foundational developmental biology to tumor trajectory modeling and treatment-informative cellular outcomes. This framework warrants attention from anyone researching cellular dynamics.”

Bridging a long-standing gap in single-cell biology

Single-cell biology research has facilitated the development of increasingly detailed mappings of cellular development. RNA velocity techniques enable researchers to estimate how cells navigate through developmental landscapes, while gene regulatory network methodologies reveal relationships among genes. However, these methods have typically functioned separately. RNA velocity approaches often neglect direct modeling of transcriptional regulation, while regulatory network methodologies do not capture cellular dynamics over time.

“Historically, cellular dynamics and gene regulation have been treated as distinct entities,” explained Professor Fabian J. Theis, Ph.D., co-senior author and Director of the Computational Health Center (CHC) at Helmholtz Munich. “RegVelo merges these areas, permitting us to investigate not just how cells transform, but also the regulatory interactions that facilitate those transformations.”

The framework jointly models splicing kinetics alongside gene regulatory relationships, allowing scientists to chart the concealed timeline of cellular development, predict transitions between cell states, and assess the outcomes of perturbing specific regulators. Practically, this shift enables researchers to pose deeper mechanistic questions than merely asking, “Where is this cell headed?” They can now inquire, “Which genes propel it there?”

Joining forces

This research underscores a profound collaboration among complementary teams. Sauka-Spengler’s lab, which transitioned from the University of Oxford to the Stowers Institute in 2022, provided a high-resolution gene regulatory scaffold for cranial neural crest development. In contrast, Theis’s group contributed computational prowess in modeling and defining developmental trajectories of individual cells through RNA velocity analyses. Together, these methodologies were synergistically integrated into a shared deep learning model that enhanced the predictive and experimental testability of developmental transitions.

What are Gene Regulatory Networks?

Video: Tatjana Sauka-Spengler, Ph.D., on decoding the cell’s instructions

Gene regulatory networks consist of the organized sets of instructions that guide a cell as it transitions from one identity to another.

“These networks function as a cascade of events,” explained Sauka-Spengler. “One group of genes activates or inhibits others, directing a cell down a particular path. This is crucial, as every cell in the body originates from identical DNA. The distinction between a skin cell, a neuron, or a muscle cell lies not in the genetic blueprint itself, but in which genes are activated, when they activate, and how they interact.”

The process can be likened to an electronic circuit. Some genes act as “go” signals, while others serve as brakes, collectively forming a code that scientists strive to decode.

How it works: a deeper dive into the findings and implications

The team employed RegVelo to identify early pigment drivers and unveiled a previously unknown regulator influencing pigment cell fate in zebrafish. The framework was applied across diverse systems, including cell cycle, pancreatic endocrinogenesis, hematopoiesis, myogenesis, hindbrain development, and zebrafish neural crest development. Across these contexts, RegVelo demonstrated performance comparable to or exceeding leading approaches in inferring latent time, velocity, terminal states, and lineage-associated drivers.

One of the standout cases arose from studying the neural crest, a developmental system responsible for generating various cell types, including pigment cells, craniofacial tissues, and components of the peripheral nervous system. RegVelo was particularly valuable here, as it identified regulators active early in developmental trajectories, even when those genes were not prominently expressed in the final cell state.

The researchers discovered that tfec operates as an early driver in pigment cell development and also recognized elf1 as a previously unknown regulator of pigment lineage fate.

Follow-up experiments, including CRISPR/Cas9-mediated knockout and single-cell Perturb-seq, confirmed both predictions, indicating that the model can formulate biologically significant hypotheses that hold up in live systems beyond mere descriptive capabilities.

“RegVelo integrates these two knowledge domains, enabling us to validate our findings,” Sauka-Spengler noted. “We can predict the key drivers behind specific cell fates and commitments and simulate perturbations to assess their downstream impacts clearly.”

This capability is crucial due to the scale of parameters researchers contend with, often involving hundreds, if not thousands, of factors. Experimentally assessing each one individually is prohibitively expensive and impractical. RegVelo helps fine-tune that search.

“Considering networks with potentially hundreds of genes, it’s unfeasible to perturb them all methodically,” Sauka-Spengler said. “Thus, we can leverage RegVelo as both an analytical and a predictive tool for future experiments.”

Computational modeling with experimental validation: a more predictive and promising combination

The research team regards RegVelo as a stride towards a more predictive branch of developmental biology, wherein computational models assist in prioritizing experiments, uncovering hidden regulators, and forecasting how cell fates might shift when gene networks are affected. RegVelo provides insight into how cells traverse state transitions and pinpoints the gene interactions steering their fate decisions.

The potential ramifications of this framework could stretch far beyond a single developmental system. A refined understanding of regulatory mechanisms could aid in identifying causes behind developmental defects and, progressively, in more accurately directing cells in therapies aimed at regenerative medicine — covering diverse applications such as cardiac muscle repair, skin graft production, and lab-grown cartilage development. Disorders related to craniofacial development, pigment cell anomalies, and broader efforts to direct stem cells or organoids towards targeted cell states are among the numerous areas where this enhanced comprehension could prove invaluable.

“Possessing a comprehensive understanding of gene regulatory circuitry, having been predicted, simulated, perturbed, and subsequently validated, presents us with a robust tool,” Sauka-Spengler concluded. “We can initiate from stem cells or naïve cells and devise new methodologies to guide them toward the desired cell types for therapeutic applications.”

The findings suggest that this framework may be extended in the future to include additional regulatory elements like chromatin, protein activity, and other multimodal metrics.

While there are current constraints — such as simplified assumptions regarding latent time, regulatory relationships, and computational demands — the study presents a compelling proof of concept. “Linking dynamic cell-state modeling directly to gene regulation facilitates a deeper exploration of mechanisms and the potential for new discoveries,” Sauka-Spengler articulated.

More information:

To read the press release from Helmholtz Munich, click here.

Additional authors include Weixu Wang, Zhiiyuan Hu, Philipp Weiler, Sarah Mayes, Marius Lange, Daniel M. Fountain, Julianna O. Haug, Jingye Wang, and Zhengyuan Xue.

Research at the Stowers Institute was backed by institutional funding to Tatjana Sauka-Spengler, Ph.D., and the Wellcome Trust Award 215615/Z/19/Z. Support for the broader collaborative project came from the European Union/ERC DeepCell project (101054957), the Wellcome Leap ΔTissue Program (9E8E84F7-8991-4D4A-A9EC), the European Union’s Horizon 2022 research and innovation program (101057775), the German Federal Ministry of Education and Research with HOPARL project (031L0289A), the DFG Graduate School of QBM (GSC 1006), the Joachim Herz Foundation, an EMBO Postdoctoral Fellowship, the Fundamental Research Funds for the Central Universities (2042025kf0022, 2042022dx0003), and the National Natural Science Foundation of China (32500725).

About the Stowers Institute for Medical Research

Founded in 1994 by Jim Stowers, the founder of American Century Investments, and his wife, Virginia, the Stowers Institute for Medical Research is a non-profit organization dedicated to biomedical research aimed at understanding life’s complexities and improving health outcomes through innovative methods targeting disease prevention, treatment, and understanding.

The Institute comprises 24 independent research programs, employing around 500 members, over 370 of whom are scientific staff, including principal investigators, technology center directors, postdoctoral researchers, graduate students, and technical staff. To learn more about the Institute, visit www.stowers.org and find more information about its graduate program at www.stowers.org/gradschool.

Media Contact:
Joe Chiodo, Director of Communications
724.462.8529
chiodo.joe@stowers.org

View original content to download multimedia: https://www.prnewswire.com/news-releases/new-ai-tool-predicts-how-cells-choose-their-future-helping-uncover-hidden-drivers-of-development-302768491.html