An AI system enables robots to run independent scientific experiments — up to 10,000 per day — potentially leading to a huge leap forward in the pace of discovery in fields from medicine to agriculture and environmental sciences.
reported today in nature microbiology, The team was led by a professor now at the University of Michigan.
Called BacterAI, this AI platform identified the metabolism of two microbes linked to oral health — with no background information to begin with. Bacteria consume a mixture of the 20 types of amino acids needed to support life, but each type requires specific nutrients to grow. The UM team wanted to know which amino acids the beneficial microbes in our mouths needed so they could promote their growth.
said Paul Jensen, an associate professor of biomedical engineering at the University of Illinois who was at the University of Illinois when the project began.
However, figuring out which group of amino acids the bacteria like is difficult. These 20 amino acids yield more than a million possible combinations, depending on whether or not each amino acid is present. However, BacterAI was able to detect the amino acid requirements for the growth of both Streptococcus gordonii and Streptococcus sanguinis.
To find the right formula for each type, BacterAI tested hundreds of amino acid combinations daily, honing their focus and changing formulations each morning based on the previous day’s results. Within nine days, it was producing accurate predictions 90% of the time.
Unlike traditional approaches that feed labeled datasets into a machine learning model, BacterAI creates its own dataset through a series of experiments. By analyzing the results of previous experiments, he makes predictions about what new experiments might give him the most information. As a result, he came up with most of the rules for feeding bacteria in less than 4,000 experiments.
“When a child is learning to walk, they don’t just watch adults walk and then say ‘OK, I get it,’ stand up and start walking. They stumble and do a little trial and error first,” said Jensen.
“We wanted our AI agent to take steps and fall, to come up with his own ideas and make mistakes. Every day, he gets a little better, a little smarter.”
Little or no research has been done on nearly 90% of bacteria, and the amount of time and resources required to learn even basic scientific information about them using conventional methods is daunting. Automated experimentation can greatly speed up these discoveries. The team ran up to 10,000 experiments in a single day.
But the applications go beyond microbiology. Researchers in any field can set up questions as puzzles that AI can solve through this kind of trial and error.
“With the recent explosion of mainstream AI over the past several months, many people are unsure what it will bring in the future, positive or negative,” said Adam Dama, a former engineer in Jensen’s lab and lead author of the study. . “But to me, it’s very clear that focused applications of AI like our project will speed up everyday research.”
The research was funded by the National Institutes of Health with support from NVIDIA.