Scientists have found a way to use machine learning to identify the environmental preferences of microbes, including bacteria, which play an essential role in maintaining ecosystems.
Genetic “fishing” techniques of recent decades have provided scientists with an extensive database of bacterial genomes, but it has been challenging to isolate and grow these microbes in the lab, so we often know little about them, except for their genetic makeup.
In any environment, there are a ton of bacteria with important ecological functions, but their environmental preferences often remain unknown. The research team drew on what scientists know about a few bacterial groups, which thrive at one particular pH or another, and then used machine learning to link those groups’ environmental pH preferences with their genetic makeup.
Machine learning is the key to unlocking the secrets of microbial life and understanding their response to environmental changes.
The work involved sorting through the genomes of more than 250,000 types of bacteria from nearly 1,500 soil, lake, and stream samples.
Understanding whether certain bacteria are most likely to thrive in acidic, neutral, or basic environments is just a first step, said lead author Josep Ramoneda, a visiting scholar at the Cooperative Institute for Research in Environmental Sciences (CIRES).
“You could use this approach to anticipate how microbes will adapt to almost any environmental change,” he said. For instance, if sea-level rise is bringing more saline water into a coastal wetland, “We can anticipate how microbes will respond to these environmental changes,” Ramoneda said.
Thanks to machine learning, we can anticipate how microbes will adapt to almost any environmental change and grow colonies of finicky bacteria we’ve never been able to grow before.
“What we found is we can make inferences about their pH preferences based on genomic data alone,” Ramoneda said. One of the finding’s most immediate implications is that it could help scientists grow colonies of finicky bacteria they’ve never been able to grow before, by giving them a first guess at what pH to use.
It can take years to figure out how to “culture” bacteria so they can be studied in the laboratory, and the machine-learning method could make that process far more efficient, said Noah Fierer, a professor of ecology and evolutionary biology at CU Boulder and a fellow of CIRES.
Agricultural and forestry experts often add live bacteria to “inoculate” growing plants with helpful communities of bacteria, Ramoneda said. Now, they may get quicker, better insight into the types of bacteria that might help restore a native prairie vs. pine forests, or to better grow corn or soybeans, by ensuring that inoculants will be adapted to the local pH.
The power of machine learning is transforming the field of microbiology, from predicting microbial adaptation to better understanding how warming will influence soil bacterial communities.
Microbes, including bacteria, are critical to the functioning of ecosystems, helping plants grow, enabling nutrient cycling in lakes, and even supporting human digestion. The new technique is a significant development that could help better understand how microbes will adapt to environmental changes and aid in growing and studying bacteria in the lab.
Next, the team plans to extend this method to identify the temperature preferences of bacteria to better understand how warming will influence soil bacterial communities, for example.
“The alternative is to try to grow them all in the lab, and that’s painful,” Fierer said. With this new technique, scientists have an opportunity to use machine learning to better understand the basic natural history of bacteria and how they respond to environmental changes.
It’s a promising new approach that could have wide-ranging implications for the study of microbiology and the environment.
