Scientific work often involves sifting through enormous amounts of data, a task that’s overwhelmingly mundane for humans but a piece of cake for artificial intelligence. A new platform dubbed BacterAI can conduct as many as 10,000 experiments per day to teach itself – and us – more about bacteria.
The human body is home to trillions of microbes, covering almost every surface inside and out. Many of them are vital to specific bodily functions, while many others make you sick. Research continues to uncover how inextricably linked our overall health is to our microbiomes, but managing and exploring the data involved remains a daunting task.
"We know almost nothing about most of the bacteria that influence our health," said Paul Jensen, corresponding author of the new study. "Understanding how bacteria grow is the first step toward reengineering our microbiome."
AI is especially good at handling huge datasets and finding patterns, so of course scientists have put it to work analyzing data on bacteria. Generally this involves feeding existing datasets into machine-learning models, but that doesn’t help for species where there simply isn’t much data available – and that’s a lot of species, considering around 90% of bacteria have received little to no study.
Researchers from the University of Michigan have now developed a new platform called BacterAI, which can study bacteria with no prior knowledge. It creates its own dataset from scratch by designing experiments for laboratory robots to run one after another, with the results of each informing the next. Eventually it can distill its findings into a set of logical rules that human scientists can understand and test further.
In a demonstration of the tech, BacterAI was put to work figuring out the metabolism of two common oral bacteria – Streptococcus gordonii and Streptococcus sanguinis. This involves identifying a specific combination of amino acids that the bugs eat from a set of 20 that support life – a task that requires sorting through over a million possible combos.
BacterAI was able to test a few hundred combinations of amino acids each day, picking the most promising combos and following those up in subsequent experiments. It was conducting up to 10,000 experiments per day, and after nine days it was able to make accurate predictions 90% of the time.
"When a child learns to walk, they don’t just watch adults walk and then say 'Ok, I got it,' stand up, and start walking. They fumble around and do some trial and error first," said Jensen. "We wanted our AI agent to take steps and fall down, to come up with its own ideas and make mistakes. Every day, it gets a little better, a little smarter."
The team hopes that BacterAI can be used to speed up discoveries about bacteria, which in turn could inform the development of new drugs or other useful molecules.
The research was published in the journal Nature Microbiology.
Source: University of Michigan
