Artificial intelligence systems could enable robots to perform up to 10,000 autonomous scientific experiments per day, potentially dramatically leaping the pace of discovery in fields ranging from medicine to agriculture to environmental science.
The team, which was reported today in Nature Microbiology, was now led by a professor at the University of Michigan.
Its artificial intelligence platform, called BacterAI, mapped the metabolism of two microbes associated with oral health. Bacteria consume some combination of the 20 amino acids necessary to sustain life, but each species requires specific nutrients to grow. The UM team wanted to know which amino acids are needed so that the beneficial microbes in your mouth can promote their growth.
“We know very little about most of the bacteria that affect our health. Understanding how bacteria grow is the first step toward restructuring our microbiome.” step,” said Paul Jensen, an assistant professor of UM biomedical engineering who was at the University of Illinois when the project was launched.
But understanding which amino acid combinations bacteria prefer is difficult. These 20 amino acids generate over a million possible combinations based on the presence or absence of each amino acid. However, BacterAI was able to discover the amino acids required for the growth of both Streptococcus gordonii and Streptococcus sanguinis.
To find the right formula for each species, BacterAI tested and focused on hundreds of amino acid combinations per day, changing combinations each morning based on the previous day’s results. Within 9 days, we were getting accurate predictions 90% of the time.
Unlike traditional approaches that feed labeled datasets to machine learning models, BacterAI creates its own datasets through a series of experiments. Predict which new experiment will provide the most information by analyzing the results of previous trials. As a result, I found most of the rules for feeding bacteria in less than 4,000 experiments.
“When a child learns to walk, they don’t just watch an adult walk, say, ‘Okay,’ get up and start walking. They fumble and do trial and error first,” says Jensen. says.
“We wanted AI agents to take a step and fall, come up with their own ideas, and make mistakes.
About 90% of bacteria have received little or no research, and the amount of time and resources required to learn basic scientific information about them using conventional methods is mind-boggling. It’s like going far away. Automated experiments can greatly speed up these discoveries. The team he performed up to 10,000 experiments in a single day.
But the applications go beyond microbiology. A researcher in any field can set a problem as a puzzle and let AI solve it through such trial and error.
Adam Dama, a former engineer at the Jensen Institute and lead author of the study, said: “But it is clear that focused applications of AI like our project will accelerate everyday research.”
This research was funded by the National Institutes of Health with support from NVIDIA.
