AI can run one million microbial experiments per year

(Nanowerk News) Artificial intelligence systems enabling robots to conduct autonomous scientific experiments—as many as 10,000 per day—potentially driving drastic leaps forward in the speed of discovery in fields ranging from medicine to agriculture to environmental science.

Reported in Natural Microbiology (“BacterAI maps microbial metabolism without prior knowledge”), the team was led by a professor who is now at the University of Michigan.

The artificial intelligence platform, dubbed BacterAI, maps the metabolism of two microbes linked to oral health — with no basic information to begin with. Bacteria consume some combination of the 20 amino acids needed to support life, but each species requires specific nutrients to grow. The UM team wanted to know what amino acids the beneficial microbes in our mouths needed to promote their growth. Paul Jensen, Assistant Professor in Biomedical Engineering at the University of Michigan and his graduate students have created an artificial intelligence agent that uses game robots to answer scientific questions. BacterAI can assign autonomous scientific experiments to robots that eventually lead to answers that would normally take humans years to answer. Their Deep Phenotyping system has completed 931,038 automated experiments since January 2020. (Image: Marcin Szczepanski, Michigan Engineering)

“We know almost nothing about most of the bacteria that affect our health. Understanding how bacteria grow is the first step to re-engineering our microbiome,” said Paul Jensen, a UM assistant professor of biomedical engineering who was at the University of Illinois when the project began.

However, figuring out which amino acid combinations bacteria prefer is tricky. Those 20 amino acids make up over a million possible combinations, based solely on the presence or absence of each one. However, BacterAI was able to find the amino acid requirements for the growth of Streptococcus gordonii and Streptococcus sanguinis.

To find the right formula for each species, BacterAI tests hundreds of amino acid combinations per day, hones its focus, and changes the combination every morning based on the previous day’s results. Within nine days, it produced accurate predictions 90% of the time.

Unlike the conventional approach of feeding labeled datasets into machine learning models, BacterAI creates its own datasets through a series of experiments. By analyzing the results of previous trials, predictions emerge as to which new experiment might provide the most information. As a result, most of the rules for feeding bacteria are known with less than 4,000 trials.

“When a child learns to walk, they don’t just see an adult walking and then say ‘Ok, I understand,’ stand up, and start walking. They fumbled and did some trial and error first,” says Jensen.

“We want our AI agents to take a step and fall, come up with ideas of their own and make mistakes. Every day, get a little better, a little smarter.

Little to no research has been done on about 90% of bacteria, and the amount of time and resources needed to study even basic scientific information about them using conventional methods is daunting. Automated experiments could drastically speed up these discoveries. The team ran up to 10,000 trials in one day.

But its applications go beyond microbiology. Researchers in any field can set up questions as puzzles for AI to solve through this kind of trial and error.

“With the recent explosion of mainstream AI over the last few months, many people are unsure about what the future holds, both positive and negative,” said Adam Dama, a former engineer at the Jensen Lab and lead author of the study. . “But to me, it was very clear that focused AI applications like ours would accelerate everyday research.”