Machine learning assists scientists in determining microorganisms’ environmental preferences.

Using machine learning, researchers have developed a method for predicting the environmental pH preferences of bacteria. The new strategy, which is being led by experts from the University of Colorado Boulder, has the potential to aid in the development of health-related probiotics, agriculture, and ecological restoration efforts.

“We know that in any climate, there are a lot of microbes with significant natural capabilities, but their ecological inclinations frequently remain obscure,” said Noah Fierer, an individual of the Helpful Establishment for Exploration in Ecological Sciences (CIRES) and teacher of environment and transformative science at CU Stone. “The goal is to use this method to learn the fundamentals of their natural history.

Lead author Josep Ramoneda, a visiting scholar at CIRES, stated, “Understanding whether certain bacteria are most likely to thrive in environments that are acidic, neutral, or basic is just the first step.” This method could be used to predict how microbes will respond to almost any change in the environment, he said. Take, for instance, the fact that rising sea levels are introducing more salty water into a coastal wetland. Ramoneda stated, “We can anticipate how microbes will respond to these environmental changes.”

“We know that there are a lot of bacteria with important ecological functions in any environment, but their environmental preferences are often unknown. The goal is to apply this approach to learn the fundamentals of their natural history.”

Noah Fierer, a fellow of the Cooperative Institute for Research in Environmental Sciences (CIRES).

Co-authors include colleagues from Canada as well as others from CIRES and CU Boulder. The new work was published today, April 28, in Science Advances.

Ecosystems cannot function without microbes, including bacteria, which facilitate the cycle of nutrients through lakes and support plant growth as well as human digestion. According to Ramoneda and Fierer, we only know their genetic makeup because it is often impossible to isolate and grow them in the lab. In recent decades, genetic “fishing” techniques have resulted in exponentially expanding databases of bacterial genomes.

Corinne Walsh, a Ph.D. candidate at the University of Colorado Boulder, works with soil samples that contain wheat plant-associated microbes. Microbial ecologists like Walsh may be able to determine the environmental preferences of bacteria by simply examining their genomes, making some laboratory work more efficient and agricultural science more successful. Credit: Cooperative Institute for Environmental Science Research (2020)

The team used machine learning to make a connection between the bacterial groups’ preferences for environmental pH and their genetic makeup, drawing on what they already knew about a few groups of bacteria that thrive at different pH levels. From nearly 1,500 soil, lake, and stream samples, the work involved sorting through the genomes of more than 250,000 different kinds of bacteria.

Ramoneda stated, “We found that we can make inferences about their pH preferences based solely on genomic data.” One of the most immediate repercussions of the discovery for scientists is that it may provide them with a first guess at the pH they should use to grow colonies of finicky bacteria they have never been able to grow before. According to Fierer, it can take years to figure out how to “culture” bacteria so that they can be studied in the laboratory. The machine-learning approach, on the other hand, might make that process much, much more efficient.

Ramoneda stated that experts in agriculture and forestry frequently add live bacteria to “inoculate” growing plants with beneficial bacterial communities. By ensuring that inoculants will be adapted to the local pH, the team can now get a better understanding of the kinds of bacteria that might help restore a native prairie versus pine forests or better grow corn or soybeans. Next, the team plans to try to learn about the temperature preferences of bacteria, which is another complicated system likely involving a lot of genes. That could help them better comprehend how, for instance, warming will affect bacterial communities in soil.

Fierer stated, “The alternative is to try to grow them all in the lab, which is painful.”

More information: Josep Ramoneda et al, Building a genome-based understanding of bacterial pH preferences, Science Advances (2023). DOI: 10.1126/

Topic : Article