Helmholtz Munich’s Fabian Theis on pushing from mapping to controlling biology with multi-scale AI models
The Virtual Cell concept aims to predict and control biological systems across scales, but recent evaluations show many models struggle even basic perturbation tasks. A $1.8B Biohub initiative seeks to close the data gap.
1 source · cross-referenced
- Fabian Theis of Helmholtz Munich argues that AI-driven systems biology must move from mapping to predicting to controlling biological systems across multiple scales.
- A $1.8B Biohub initiative aims to generate high-precision, multi-scale biological data to enable predictive modeling from molecules to organisms.
- Recent evaluations, including the 2025 Arc Virtual Cell Challenge, found that many models fail to beat simple baselines for perturbation prediction, highlighting the need for better data and metrics.
- Theis’s labs are developing algorithms for multi-omics, time-series, and clinical data integration to drive precision medicine.
Fabian Theis, a leader in systems biology at Helmholtz Munich, is advancing the "Virtual Cell" concept—a framework to predict and ultimately control biological systems across multiple scales, from molecules to organisms. His work emphasizes moving beyond static mapping to dynamic, predictive modeling that can guide interventions in medicine and biology.
A $1.8B initiative by Biohub, NIH, DOE, Google, and Meta aims to generate large-scale, high-precision biological data to fuel these models. The effort focuses on new imaging and measurement technologies, including lab-in-the-loop automation and advanced microscopy, to capture co-registered, multi-modal data at unprecedented scales.
Recent evaluations, such as the 2025 Arc Virtual Cell Challenge, have revealed that many sophisticated models struggle to outperform simple baselines—even constant or linear models—in predicting cellular responses to perturbations like gene editing or drug exposure. This has raised questions about the effectiveness of current approaches and the adequacy of existing datasets.
Theis argues that perturbation prediction may not yet be a sufficiently challenging benchmark for complex algorithms. Instead, he advocates for tackling larger, multi-scale problems, such as predicting how perturbations propagate across scales—from molecules to cells, tissues, organs, and entire organisms. Addressing these "scale interfaces" requires both new data and more advanced modeling techniques.
His teams at Helmholtz are already working on these challenges. The Computational Health Center integrates AI with molecular data, medical imaging, genomics, and clinical records to develop predictive models for precision medicine. Meanwhile, the Institute of Computational Biology focuses on statistical and dynamical models that bridge single-cell time series with population-scale omics data.
The conversation highlights a pivotal moment in systems biology: while AI is enabling the tackling of larger problems, the field still lacks critical data and robust evaluation metrics. Initiatives like Biohub’s $1.8B push aim to close these gaps, but success will depend on whether the resulting data and models can deliver on the promise of controlling biology.
- Nov 11, 2026 · Latent Space — swyx
OpenAI demonstrates voice steering, cloud handoffs, and personal agents for Codex at AI Engineer Code Summit
Trust79 - Nov 11, 2026 · The Verge — AI
OpenAI to launch verified AI Business Agents in ChatGPT for US e-commerce and service businesses
Trust79 - Nov 10, 2026 · Latent Space — swyx
NaiveAI releases Naive-N0.5-Flash coding model with ultrafast token throughput
Trust74