Scientists Create AI ‘Virtual Mouse’ to Replace Animal Experiments

For years, researchers developing nanomedicines have injected mice with different nanoparticles to discover where those materials travel inside the body.

The particles may accumulate in the liver, kidneys, lungs or spleen. They may cross protective barriers, damage organs or fail to reach their intended target. Each variation in size, coating, shape or surface charge can produce a different result. Finding the right combination has traditionally meant using more mice.

Now, a researcher in Switzerland has created a virtual mouse that can conduct many of these experiments without using another living animal.

Testing Nanomedicine Without Mice

Nanomedicine uses extremely small materials to diagnose, prevent or treat disease. Some nanoparticles are small enough to cross barriers within the body that conventional medicines struggle to penetrate. This could make them particularly useful for treating brain tumours. The blood-brain barrier protects the brain by blocking many substances, including most drugs. Carefully designed nanoparticles could cross that barrier and deliver chemotherapy directly to a tumour. But first, researchers need to understand how each nanoparticle behaves.

Jimeng Wu, a doctoral researcher at the Swiss Federal Laboratories for Materials Science and Technology, developed an artificial intelligence system that predicts how nanoparticles will move through a mouse’s body.

The system combines a virtual representation of the body with machine learning. Researchers enter measurable characteristics such as the particle’s size, coating and surface charge. The model then predicts how much of the material will reach different organs and how long it will remain there. Unlike conventional computer models, which often need to be recalibrated for each individual material, Wu’s system can adapt its calculations to different nanoparticles.

Researchers can therefore screen potential materials before manufacturing them, let alone injecting them into mice.

Built From Existing Experiments

The virtual mouse was trained using data from 18 previously published experiments on living mice. This does not erase the harm behind the original data. Those mice were still used. But the model can extract more information from experiments that have already happened instead of repeatedly reproducing them.

The researchers tested the system across five types of non-dissolvable nanoparticles. It performed particularly well when predicting broad indicators such as how much material reached different organs. Size, coating and surface charge were among the most important factors influencing the predictions.

The model is not finished.

Eighteen studies represent a small dataset. Many published papers also fail to describe the nanoparticles in enough detail to be useful. Some of the model’s predictions were considerably stronger than others, particularly when estimating how concentrations changed over time. The researchers say more data, further verification and more complex modelling will be needed. That is how science works. A new method does not need to be perfect before researchers begin developing and using it. Animal experiments certainly were never perfect.

From Virtual Mice to Virtual Humans

Wu’s longer-term goal is not simply to create a better simulation of a mouse. She plans to transfer the same principles into a virtual human model.

The current system predicts nanoparticle distribution in a mouse’s liver, kidneys, lungs and spleen. A human version could help researchers investigate how nanoparticles interact with organs including the brain.

That could make the research more directly relevant to the people medicines are intended to treat. Mice are not miniature humans. Differences between species affect metabolism, immune responses, organ function and disease progression. An experiment can produce consistent results in mice and still fail when a treatment reaches human trials.

Replacing a living mouse with a virtual mouse removes immediate animal use. Moving from mouse biology to human biology addresses the deeper scientific problem. The aim should not be to recreate animal experimentation on a computer forever. It should be to develop research designed around humans from the beginning.

AI Is Already Reducing Animal Use

Wu’s model is not an isolated experiment.

Researchers in Germany have developed another generative AI system called genESOM. It learns from small experimental datasets and creates synthetic data points that reflect the patterns found in real results. Its developers estimate that the system could reduce the number of animals used in some preclinical drug experiments by between 30 and 50%.

The UK government has also published a roadmap intended to accelerate the replacement of animal experiments. The plan supports AI analysis, organs-on-chips and 3D-bioprinted human tissues. It set targets to end certain skin and eye irritation experiments by the end of 2026, end mouse experiments used to measure Botox potency by 2027, and reduce pharmacokinetic experiments on dogs and primates by 2030.

The technology is advancing. Governments are investing. Researchers are demonstrating that alternatives can be faster, cheaper and more relevant to human health.

So why are animals still being used?

The Problem Is Scientific Culture

Animal experimentation is not sustained by scientific evidence alone. It is also sustained by habit, institutional authority and a peer-review system that continues to treat animal use as the default. A survey of 100 researchers in India found that 56% had been asked by peer reviewers to add animal experiments to research that initially contained none.

Around one in five had conducted animal experiments before being asked because they expected reviewers to demand them. Researchers who resisted faced consequences. More than half reported manuscript rejection or withdrawal, while 59% said the pressure pushed their work into lower-impact journals. Many also believed excluding animal experiments weakened their chances of receiving funding.

Animals were not always being used because an experiment required them. They were being used because researchers feared their work would not be published without them.

That is not scientific necessity. It is institutional coercion.

When Science Becomes an Excuse

Scientific authority can also make people more willing to accept harm.

In one experiment, participants were asked to administer increasingly toxic doses to what they believed was a living fish as part of Alzheimer’s research. The fish was actually a robotic model.

More than half continued until the dose supposedly had a 100% chance of killing the fish.

A second experiment found that participants encouraged to think positively about science were more likely to continue administering harmful doses despite seeing apparent signs of distress.

The researchers concluded that identification with scientific authority can make people more accepting of “instrumental harm”. Hurting someone becomes easier to justify when an institution promises that something valuable might result.

But scientific progress does not make exploitation morally neutral.

Science is a method for understanding the world. It is not permission to do whatever researchers find convenient.

The Replacement Is Already Underway

An estimated 125 million animals are used in laboratories around the world each year.

They are poisoned, infected, genetically manipulated, restrained and killed because animal experimentation has been built into research, funding and regulation over generations.

The virtual mouse shows that another system can be built.

This particular model remains limited. It needs more data. It needs independent verification. It cannot yet replace every experiment involving nanoparticles.

But it can already help researchers reject unsuitable materials before they are manufactured or administered to living mice. It can reduce duplication. It can make development faster and cheaper. Its principles may eventually support research based directly on human biology.

The obstacle is no longer a complete absence of alternatives. It is a scientific establishment that continues to demand animal experiments even when researchers offer something better.

Mice are not laboratory equipment. They should never have been treated as disposable containers for human data.

A virtual mouse cannot suffer.

That alone makes the choice obvious.