Scientists from the Stowers Institute for Medical Research, Helmholtz Munich, and the University of Oxford have developed RegVelo, a new artificial intelligence framework designed to predict cell fate and identify the genetic regulators driving development. Published in the journal Cell, this tool bridges a long-standing gap in single-cell biology by combining two previously separate methods: RNA velocity, which estimates how cells change over time, and gene regulatory network analysis, which maps relationships among genes. By integrating these approaches, RegVelo allows researchers to simulate cellular transitions and pinpoint specific genes that steer cells toward their final identities, effectively enabling virtual experiments that reduce the need for costly laboratory trials. The study focused primarily on neural crest cells, a versatile group of embryonic cells that contribute to the face, heart, nervous system, and pigment cells. In zebrafish models, RegVelo successfully identified tfec as an early driver of pigment cell formation and revealed elf1 as a previously unknown regulator of pigment cell fate. These computational predictions were subsequently validated through experimental methods, including CRISPR/Cas9-mediated knockouts and single-cell Perturb-seq, confirming the model's ability to generate accurate biological hypotheses. Researchers noted that many critical regulatory elements are lost when analyzing only the final cell state; RegVelo captures these dynamic events throughout the entire developmental timeline. Tatjana Sauka-Spengler, a senior author of the study and investigator at the Stowers Institute, emphasized that understanding the initiating elements of development is crucial for regenerative medicine. She explained that if scientists can reproduce these early instructions in vitro, they can generate specific cell types for therapies, such as repairing heart muscle or growing skin grafts. Alejandro Sánchez Alvarado, the President and Chief Scientific Officer at Stowers, highlighted that the framework's use of deep learning allows for the inference of likely developmental paths and the experimental testing of those dynamics. The collaboration merged high-resolution gene regulatory data from Sauka-Spengler's team with computational expertise from Fabian J. Theis's group at Helmholtz Munich. This union created a robust model capable of predicting outcomes across various biological systems, including the cell cycle, pancreatic development, blood formation, and tumor trajectories. Theis noted that while cellular dynamics and gene regulation were historically modeled separately, RegVelo now allows scientists to ask mechanistic questions about which specific interactions drive cellular changes. Beyond developmental biology, the framework offers significant potential for studying diseases and advancing cell therapy. By simulating the effects of perturbing specific genes, RegVelo serves as both an analysis and a screening tool, helping researchers narrow down the hundreds of potential factors involved in complex gene networks. While the current model includes certain simplifying assumptions regarding latent time and computational cost, it represents a major step toward predictive developmental biology. Future iterations may incorporate additional data layers, such as chromatin and protein activity, to further refine the understanding of gene regulatory circuits and their role in health and disease.