Engineering Cell Fate: Towards a Foundation Model for Virtual Cell Signaling

We are building a future where biology is no longer destiny, but design.

January 7, 2026

The Challenge
Modern biology struggles to predict and control cellular behavior because cell signaling is complex and context-dependent. Current methods rely on slow, empirical trial and error.

The Cellular Intelligence Solution
Cellular Intelligence is building the first Universal Virtual Cell-Signaling Model: a foundation model capable of predicting how any cell in any state changes in response to external signals.

The Competitive Advantage

• Unrivaled Data Scale: Utilizing a proprietary capsule-based platform, Cellular Intelligence generates massive, context-rich datasets—scaling to millions of unique perturbation conditions—to solve the problem of context dependence.
• Static vs. Dynamic States: While others profile cells in fixed states, we use human stem cells to decode the combinatorial signaling logic that determines cellular behavior and ultimately cell fate,

Core Architecture
Our framework is built on a synergistic feedback loop between massive-scale data generation and predictive modeling. This proprietary capsule data engine covers the astronomical search space of cell signaling, distilling it into the context-rich, high-fidelity datasets required to train transformer models to learn the fundamental "grammar" of cellular signaling.

Translational Impact
By transforming biology into a predictive engineering discipline, Cellular Intelligence enables in silico control of cellular behavior, with applications ranging from rational protocol design for regenerative medicine to context-specific drug effect prediction and systematic disease modeling.

1. A New Era of Cellular Control: From Empirical Biology to Predictive Control

A major bottleneck in modern medicine is the inability to predict how different cells respond to signals. We are replacing slow, manual experimentation with a predictive model that handles this complexity, accelerating the path to life-saving therapies.

A fundamental challenge in modern biology is that of precise, engineered cellular control. Cells possess their own language for communicating with each other—cell signaling—which directs core biological processes like development and is frequently dysregulated in disease.

Cells possess their own language for communicating with each other.

Remarkably, biology achieves this complexity using a surprisingly concise vocabulary: only around 20 fundamental molecular signaling pathways have been identified to date. It is the combinations and orders in which they are used that underlies how such a small number of pathways can give rise to the staggering diversity of human cell types and states.

However, despite decades of effort, we have not yet deciphered the grammar of this language. Today, the effects of a given signal are largely determined through an empirical, trial-and-error process.

1. Combinatorial Complexity: The sheer number of signal combinations limits systematic experimental dissection.
2. Context Dependence: The effect of a signal depends heavily on the state of the cell prior to receiving it.

The Human Cost of Technical Limitations. The failure to decode the logic of cell signaling is not just a scientific bottleneck, but a systemic barrier to progress and, consequently, a delay in saving lives.

Cellular Intelligence’s Vision

This white paper outlines Cellular Intelligence's solution to the challenge of predicting and controlling cellular behavior: the construction of the first Universal Virtual Cell-Signaling Model.

2. What is a Virtual Cell-Signaling Model?

The virtual cell-signaling model acts as a computational twin, using a cell's initial state to accurately predict how it responds to signals.

In essence, a virtual cell-signaling model is a predictive map from an initial cell state and an external signal to the cell's future state. Formally, it can be seen as a function:

f(initial cell state, signal) → future cell state

3. A Unique Approach to Building the "Virtual Cell"

Most AI models in biology fail because they are trained on limited data—like trying to learn a language by reading just one book. Cellular Intelligence uses stem cells to generate massive, proprietary datasets that cover the entire 'tree of life,' capturing how cells behave in every possible context.

Biology has run into a complexity barrier that is now blocking progress.

Advantages of Our Approach

• Exponentially Scalable Data Collection via Capsule Technology: Our proprietary capsule-based context generation system allows us to interrogate an exponentially expanding set of signaling factors combinations and cell states.
• Active Learning Loop and Data Augmentation: Building a predictive model is only half the battle—the other half is using it intelligently to accelerate learning.
• Translational Relevance by Design: From day one, Cellular Intelligence aligned its data and model to real-world therapeutic contexts.

4. Competitive Landscape: Other Approaches and How We Differ

Many players in the field are advancing complementary components of a broader “virtual cell” vision. Cellular Intelligence focuses on a distinct challenge: understanding the cell’s built-in control systems (signaling), allowing us to predictably guide cells towards therapeutic outcomes.

Distinct Players

• Chan Zuckerberg Initiative: Focus on observational mapping rather than predictive control.
• Genentech/Roche: Primarily hypothesis generation—identifying analogs of a disease state.
• Xaira Therapeutics: Mapping the wiring vs. operating the controls.

5. Applications and Impact

1. Rational Design of Cell Differentiation Protocols: Developing protocols for regenerative medicine becomes systematic with a virtual signaling model.
2. Context-Specific Drug Response Prediction: The model can simulate how specific cell types respond to treatments in varying conditions.
3. Genetic Disease Modeling and Phenotypic Rescue: Enables discovery of small-molecule signals that compensate for genetic defects.
4. Interpreting Disease as Signaling Network Failure: Simulates abnormal responses in diseased states, aiding target identification.
5. Uncovering Hidden Biology and New Pathways: Systematic exploration can yield novel insights and potential new therapies.

6. Progress and Validation to Date

Cellular Intelligence has executed some of the largest combinatorial signaling screens in history, demonstrating the model’s processing capabilities and biological relevance.

Milestones Achieved

• Unprecedented Data Scale and Diversity: Built a dataset of sequential cell-signaling responses.
• Scaling to 1 Million Conditions: A single run confirmed the biological breadth of the model.
• Technical Validation: Achievements confirming the reliability of experimental methodologies.

7. Architecture and Training Approach

We adapt advanced neural network architectures to learn complex signaling behaviors.

Core Learning Problem

Our model is trained to approximate a transition function for cellular state, accounting for context-dependent variability.

8. Roadmap and Milestones

Cellular Intelligence is executing on a clear roadmap, progressing from single-signal predictions to a universal simulation engine capable of real-world applications.

9. Conclusion: From Observation to Engineering

Cellular Intelligence replaces decades of empirical guesswork with a predictive engine for therapeutic innovation. This transition represents profound opportunities in the medical and therapeutic landscape.

10. References

Adduri, A. K., et al. (2025). Predicting cellular responses to perturbation across diverse contexts with STATE. bioRxiv. Link
Tang, L. (2025). The virtual cell. Nature Methods. Link
Heimberg, G., et al. (2025). A cell atlas foundation model for scalable search of similar human cells. Nature. Link
Huang, A. C., et al. (2025). X-Atlas/Orion: Genome-wide Perturb-seq Datasets. bioRxiv. Link
Pearce, J. D., et al. (2025). A Cross-Species Generative Cell Atlas Across 1.5 Billion Years of Evolution. bioRxiv. Link
Rood, J., et al. (2024). Toward a foundation model of causal cell and tissue biology. Cell, 187(17), 4520-4545. Link
Rito, T., et al. (2025). Timely TGFβ signalling inhibition induces notochord. Nature. Link
Wagner, A., et al. (2016). Revealing the vectors of cellular identity with single-cell genomics. Nature Biotechnology. Link
Xu, Y., et al. (2023). A single-cell transcriptome atlas profiles early organogenesis in human embryos. Nature Cell Biology. Link