Controlling Biology - Fabian Theis, Helmholtz Munich
About
Fabian Theis of Helmholtz Munich, whose group builds algorithms for multi-scale, multimodal and time-resolved single-cell data, discusses where virtual-cell models stand. He argues that perturbation prediction on existing datasets is not a hard enough problem to justify complex models, given findings that simple baselines often match sophisticated ones and that common evaluation metrics are poorly calibrated, and that the field should instead model how molecules, cells, niches, tissues and organs affect one another using large co-registered datasets such as those Biohub plans to collect. Topics include moving from cell atlases to prediction to steering the next experiment, closing the lab-in-the-loop, the bitter lesson in biology, the business of virtual cells and training the next generation of computational biologists.