Our Mission

Honest data.
Meaningful progress.

We believe eliminating research animal use is one of the most important goals of science. Achieving it requires honesty about how hard the problem actually is.

The Measurement Crisis

The Gold Standard
is scored by hand.

Mouse and rat behavioral assays are the Gold Standard last checkpoint before clinical trials for good reason. They capture efficacy signals and flag dangerous or lethal side effects that no in-vitro system can. However, the data from existing scientific equipment are typically scored by hand, or else require human supervision to ensure that the semi-autonomous computer scoring is accurate. The human element is not only financially expensive, the variability of this outdated scoring method means that lots of animals are needed.

Virtual Labs is solving this with MouseOS, a cloud-based animal behavior analysis, data management and experimental collaboration platform that's free to any lab, anywhere in the world. You pay only the compute cost for each experiment.
Zero software license fees.
Zero contracts.
Zero fuss.

Two high unmet needs. One platform. Your experimental data is safe, secure, and always owned by you. And every experiment analyzed by MouseOS delivers better behavioral science today — and contributes to the training dataset that makes virtual lab mice possible. And for those who need the best data available, our VirtualLabs Hardware product lineup delivers unparalleled data quality and user experience for less than $10k per device.

The Founding Vision

We're building the
virtual lab mouse.

Living, breathing laboratory animals can, and should, be replaced with more humane alternatives. The responsible way to do this is by creating high-fidelity digital twins of mice and rats, that are indistinguishable from the real thing. This is not a distant hope. It is the specific, scientifically grounded conviction that led us to found Virtual Labs.

But getting there requires something the field does not yet have: large, standardized and well annotated behavioral datasets from thousands of experiments run under reproducibly controlled conditions. Here's the beautiful part — the stepping stone to that data is the platform above: every experiment analyzed by MouseOS contributes to the training dataset that makes virtual lab mice possible.

The Long Game

The path to virtual animals
runs through real data.

Stanford's 2025 digital twin of the mouse visual cortex required more than 900 minutes of neural recordings to train.1 MouseGPT required 42 million annotated behavioral frames.2 The International Brain Laboratory's decision-making dataset required 5 million recorded choices across 140 mice, 7 laboratories, and 3 countries.3

The pattern is consistent: meaningful computational models of brain function emerge from massive, high-quality, standardized empirical datasets. Those datasets don't yet exist for behavioral science. We're building the infrastructure to create them — one experiment at a time.

See Our Technology

Our Method

Honesty is
our method.

The mammalian brain remains one of the least understood objects in the known universe — approximately 86 billion neurons, an estimated 100 trillion synaptic connections, and emergent properties that cannot be reduced to a cell assay or a computer simulation. Not yet. Despite decades of extraordinary progress, a complete computational model of even the mouse brain does not exist, and the gap between where the field is and where it needs to be to replace behavioral animal models is real and large.

This is not a counsel of despair. It is a call for honesty — which is the only foundation on which real progress can be built.

Read our full position on animal research and reduction →

[ Data Infrastructure ]

Step One

Any assay.
Dramatically fewer animals.

Virtual Labs pairs purpose-built capture hardware — beginning with the PolarEye camera — with MouseOS, the free analysis platform that quantifies behavior from video of any assay: open field, home-cage behavior, gait analysis, mazes, swim tests, and beyond — whatever camera captured it.

These assays are labor intensive, prone to inter-rater variability, and expensive to run at scale. Automating them with consistent hardware, optimal lighting, and high-fidelity video analysis dramatically improves data quality and reduces the number of animals needed to reach statistically valid endpoints. That is a concrete, achievable, meaningful reduction in animal use. We're doing it today.

This is step one. Not the final destination. But a real step — grounded in science, not wishful thinking.

References & Further Reading

1 Isbister, J. et al. (2025). Towards a digital twin of the mouse brain. Stanford Medicine News. med.stanford.edu — Stanford's AI model of the mouse visual cortex required 900+ minutes of neural recordings to train a single sensory region; illustrates the scale of data required for computational brain models.
2 Xu et al. (2025). MouseGPT: A Large-scale Vision-Language Model for Mouse Behavior Analysis. bioRxiv. doi:10.1101/2025.03.27.645630 — Large-scale multimodal modeling of mouse behavior; illustrates the annotation scale behind behavioral foundation models.
3 The International Brain Laboratory et al. (2021). Standardized and reproducible measurement of decision-making in mice. eLife, 10:e63711. doi:10.7554/eLife.63711 — The IBL's standardized decision-making pipeline: 140 mice trained across seven laboratories in three countries, demonstrating that multi-site behavioral data can be made reproducible at scale.