Latent Space: The AI Engineer Podcast

Virtual cells move from mapping biology to steering experiments

Controlling Biology - Fabian Theis, Helmholtz Munich

The episode examines whether computational models can reliably predict biological interventions and close the loop between AI and laboratory experiments.

3 key takeaways
  1. 1Virtual cells are evolving from descriptive maps into systems that predict how cells respond to diverse perturbations.
  2. 2Organoids and spatial models expand biological realism while exposing gaps in current representations of multicellular systems.
  3. 3Strong baselines, proprietary data, and interdisciplinary teams may matter as much as larger foundation models for commercial impact.

Don't miss

Fabian Theis describes LabCompass’s vision for selecting biological interventions and repeatedly steering experiments with computational models.

The brief

Fabian Theis traces virtual cells from maps of cellular states toward predictive systems that model perturbations and guide the next experiment.

The discussion expands the problem from gene knockouts to any input that changes cell state, including signaling, disease, time, media, and human genetics.

Organoids add multicellular realism, but their missing immune systems and incomplete physiological context make spatial representation and biological validation harder.

Theis argues that foundation models need careful objectives and honest baselines: linear methods remain strong for easy or noisy tasks, while nonlinear models help with complexity.

LabCompass represents the episode’s central shift: an inverse generative model designed to select interventions and help close the loop between computation, experiments, and AI-driven design.

The conversation ends on the practical constraints—proprietary perturbation data, interdisciplinary training, European commercialization, and the culture required for repeated model-experiment cycles.

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.