Controlling Biology: Fabian Theis and the Frontier of Virtual Cells

Virtual Cells: From Concept to Frontier

Fabian Theis is a key figure in the Virtual Cell concept and its various initiatives. Based on the original source material, it is no exaggeration to say that his lab has been pushing the algorithmic frontier to understand multi-scale, multi-modal data. The virtual cell is not a single technology but a vision: integrating information from the molecular to the cellular to the tissue level through computational models to predict and regulate the behavior of biological systems. Theis's work at Helmholtz Munich is a core driver of this vision.

Single-cell sequencing is one of the key technologies underpinning the virtual cell vision. Traditional bulk sequencing mixes thousands of cells together, yielding only population-average signals and masking heterogeneity between cells. Single-cell RNA sequencing (scRNA-seq) isolates individual cells through microfluidic partitioning, microwell plates, or droplet encapsulation, then amplifies and sequences their transcriptomes to obtain each cell's independent gene expression profile. The technology was first demonstrated in 2009 and rapidly became widespread after 2015 with the commercialization of platforms such as 10x Genomics. It enables researchers to identify rare cell types, reconstruct developmental trajectories, and characterize cellular composition within tumor microenvironments. For virtual cells, single-cell data provides the foundational coordinates for constructing a cell state space, as well as an indispensable source of large-scale training sets for training predictive models.

Algorithmic Challenges of Multi-Scale, Multi-Modal Data

The core difficulty of virtual cells lies in the complexity and heterogeneity of the data. Biological systems span multiple scales—gene expression, protein interactions, cell types, tissue microenvironments—and present in multiple modalities (such as single-cell sequencing, spatial transcriptomics, and imaging data). Theis's lab works on developing algorithms to integrate these data and build computable cell models. This is not merely a data science problem but a transformation in how biology is understood: moving from descriptive to predictive and interventional.

Spatial transcriptomics is a class of technologies that has emerged in recent years, measuring gene expression while preserving the spatial coordinates of tissue. Unlike single-cell sequencing of dissociated tissue, spatial transcriptomics can reveal the in situ positional relationships of cells within tissue—for example, the spatial proximity of tumor cells and immune cells, or gene expression gradients across different cortical layers. Representative platforms include in situ hybridization-based MERFISH and sequencing-based Slide-seq and Visium. Spatial data adds a crucial tissue-level dimension to virtual cell models: expression profiles alone cannot distinguish "which cells are together," whereas spatial information enables models to learn cell-cell communication, microenvironmental signals, and structural constraints. Integrating spatial data with single-cell data is one of the core challenges of Theis's lab's multi-modal algorithmic work, requiring dedicated alignment, deconvolution, and graph modeling methods.

Controlling Biology: From Understanding to Intervention

The title of Theis's work, "Controlling Biology," points to a deeper goal: not only to understand biological systems but to control them. This means algorithms must not only describe cell states but also predict responses to perturbations and then design intervention strategies. This has profound implications for drug discovery, synthetic biology, and precision medicine. If virtual cells can simulate perturbation experiments in silico, they will greatly accelerate biomedical research and reduce reliance on wet-lab experiments.

Perturbation response prediction is the concrete computational embodiment of controlling biology. The basic setup is: given a cell's baseline expression profile and an applied perturbation (such as gene knockout, drug treatment, or cytokine stimulation), predict the change in cell state after perturbation. In recent years, technologies such as Perturb-seq have combined CRISPR gene editing with single-cell sequencing, enabling systematic measurement of transcriptomic responses to thousands of gene perturbations in a single experiment, providing large-scale data for training such predictive models. If virtual cells can accurately predict perturbation effects, they can "rehearse" experiments in silico, screening the most promising drug targets or gene intervention combinations, thereby significantly reducing the trial-and-error cost of wet-lab experiments. This capability is also a key marker of the virtual cell's transition from a descriptive model to an interventional tool.

Open Science and Community Collaboration

Virtual cell initiatives typically emphasize open science and community collaboration. As a key participant, Theis's work is often accompanied by data sharing and open-source tools. This model helps build consensus and standards and advances the entire field. Although the original source material does not provide specific tool names or datasets, it can be inferred that his lab's algorithmic contributions are publicly available and serve the broader research community.

Future Outlook

Virtual cells are still in an early stage of development, facing challenges such as data standardization, model interpretability, and cross-scale integration. Theis's continued efforts indicate that this direction is moving from proof of concept to practical application. As algorithms and data accumulate, controlling biology may become the paradigm for the next generation of biomedical research.

One-line Summary

Fabian Theis advances virtual cells and multi-scale algorithms toward a predictable, interventional controlling biology.

Paragraph Summary

The article introduces Fabian Theis's core driving role in the Virtual Cell concept. The virtual cell is not a single technology but a vision of integrating multi-scale, multi-modal data from the molecular, cellular, and tissue levels through computational models to predict and regulate the behavior of biological systems. Theis's lab works on developing algorithms that integrate heterogeneous data such as single-cell sequencing, spatial transcriptomics, and imaging, moving biology from descriptive to predictive. The deeper goal is "controlling biology": not only understanding systems but also predicting perturbation responses and designing intervention strategies for drug discovery, synthetic biology, and precision medicine. Open science and community collaboration are important features of the initiative. The article also notes that virtual cells still face challenges such as data standardization, model interpretability, and cross-scale integration.