An international research team involving Helmholtz Munich, the Technical University of Munich (TUM), and the University of Oxford has developed an AI-based model that can analyze developmental pathways of individual cells and simulate targeted interventions in gene regulation. The model named "RegVelo" has now been published in the journal Cell.

The researchers aim to better understand how immature cells develop into specialized cell types such as nerve, blood, or pigment cells during development. Modern single-cell technologies already provide detailed information about which genes are active in individual cells. However, it has been challenging to understand which regulatory genes control these processes and how changes in individual genes affect cell development.

This is where RegVelo comes in: The AI model combines data-driven analyses with biological network models and recognizes not only which genes become active or inactive but also how genes mutually regulate each other. This allows for simulating developmental pathways of cells and predicting possible consequences of genetic interventions.

The researchers tested the model, among other things, on the development of neural crest cells in the zebrafish embryo. RegVelo identified known regulatory mechanisms and additionally discovered previously little-studied candidates. Experimental tests confirmed central predictions of the model.

The basis of RegVelo is the so-called RNA velocity method, which can deduce developmental directions of individual cells from gene activities. The new model expands this approach with gene regulatory networks, thereby enabling significantly more precise predictions of biological processes.

In the long term, the researchers see this as an important step towards virtual cell models that could help to better understand disease mechanisms and identify new therapeutic approaches. The work also demonstrates the potential of hybrid AI systems that combine data-driven methods with biological expertise.