AI System Cuts Animal Testing in Drug Discovery by 50 Percent
Researchers at Goethe University Frankfurt have developed an artificial intelligence platform that reduces the number of animals needed in early-stage drug testing by up to 50 percent while maintaining scientific validity.
The system, called genESOM, addresses a persistent tension in preclinical research: minimizing animal use for ethical reasons while ensuring studies include enough subjects to produce statistically reliable results.
How genESOM Works
Jörn Lötsch, a data scientist and clinical pharmacologist at Goethe University Frankfurt, and Alfred Ultsch, a computer scientist at Philipps University Marburg, built genESOM using a network of artificial neurons trained on existing experimental data.
The system learns the internal structure of a dataset, then generates additional data points that behave as though they came from real experiments. This effectively simulates a larger group of animals than was actually used in the original study.
A key challenge with generative AI in science is error inflation-when random noise or insignificant variations amplify into false-positive findings. genESOM separates the learning phase from the data synthesis phase, allowing researchers to introduce an artificial error signal and measure how it spreads through generated data. The system stops generating synthetic data before scientific validity breaks down.
Test Results in Multiple Sclerosis Research
The researchers tested genESOM using data from a mouse study of multiple sclerosis. The original experiment used 26 mice divided into three treatment groups.
When Lötsch and Ultsch reduced the dataset to 18 mice (six per group), the treatment effects disappeared and results lost statistical significance. After expanding the reduced dataset with genESOM-generated data, the treatment effects reappeared at the same significance level as the original study without introducing false positives.
More complex deep-learning neural networks failed to achieve the same result.
Limitations and Future Use
Lötsch cautioned that genESOM cannot replace real experimental data entirely. "If too few animals are included in an experiment and the number is then simply supplemented using generative AI, the experiment could quickly become scientifically worthless due to the amplification of random findings," he said.
After testing multiple datasets, Lötsch concluded: "With genESOM, the number of animals used in exploratory research can be reduced by 30 to 50 percent while maintaining scientific validity."
The researchers believe the technology could substantially reduce animal experiments across preclinical research. For professionals in science and research, understanding how AI can optimize experimental design is increasingly relevant to your work. AI for Science & Research courses cover these applications in laboratory optimization and scientific discovery.
Your membership also unlocks:
