Controlling Biology - Fabian Theis, Helmholtz Munich
November 6, 2026
AI Summary
5 min readSingle-cell biology has entered a phase where scientists can move beyond simply mapping cell types and predicting how cells respond to perturbations. They can now begin to steer experiments, using generative models to ask what intervention is needed to reach a desired cellular state. Fabian Theis, director of the Computational Health Center at Helmholtz Munich, describes this as the next frontier—one that requires not only better models and bigger datasets but also a cultural shift in how computational and wet-lab scientists collaborate.
What a Virtual Cell Actually Means
The phrase "virtual cell" gets thrown around a lot, but Theis is careful to define what it entails. At its core, a virtual cell is a model that can predict how a cell will behave under any kind of perturbation—a drug, a genetic knockout, a change in its environment. This is not just a static map of cell types; it is a dynamic input-output system. Theis frames it as a world model problem: given a cell state x and an action a (the perturbation), the model must predict the next state x'.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:00) **Steering, Not Just Mapping** - Fabian introduces the shift from mapping cell atlases to guiding experiments, and the concept of "Lab Compass," an inverse generative model.
- 2 (02:09) **The Promise and Hype of Virtual Cells** - RJ asks if we are about to see renewed pharma interest in virtual cells, citing the BioHub initiative.
- 3 (05:51) **What is a Perturbation and Why Does Scale Matter?** - Fabian explains what a perturbation is and how the explosion of perturbation data is unlocking new modeling capabilities.
- 4 (08:13) **Biology 101: From Transcriptome to Perturbation** - Fabian provides a basic explanation of the transcriptome and perturbations for a non-expert audience.
- 5 (10:49) **The Rich Space of Perturbations** - Fabian expands the definition of a perturbation beyond CRISPR to include all types of cellular inputs.
- 6 (12:58) **Organoids: The Next Frontier in Perturbation Modeling** - Fabian introduces organoids as a more complex and realistic system for studying perturbations.
- 7 (15:56) **The Bitter Lesson vs. Inductive Bias in Biology** - Brandon asks how much we should rely on big data versus built-in scientific knowledge (inductive bias) in models.
+ Full timestamped outline available in the app
Show Notes
Fabian Theis has been one of the key players in the various iterations of the Virtual Cell concept and initiatives. It is not hyperbole at all to say that his lab has been pushing the frontier of algorithms to understand multi-scale, multi-modal (”mult...