Machine learning helps researchers identify underground fungal networks

Justin Stewart left for Mount Chimborazo in August 2022 to collect fungal samples from the Ecuadoran volcano at an elevation of 4,000 meters, or about 13,000 feet. Given that vegetation would be sparse at that altitude, Stewart says he didn’t expect to find enough plant roots underground that would support mycorrhizal fungi, the species he had set out to sample.

Once up there, Stewart and his team started digging holes to collect samples. But much to their surprise, he later said, it was filled with roots and plant systems. During lab analysis later, they identified 12 species of arbuscular mycorrhizal fungi — the essential interface that facilitates the transfer of water and nutrients from soil to plants via the root system.

“This is an area where we had predicted high biodiversity,” Stewart told Mongabay in a video interview. “To see that there are actual root systems there, and fungi, was so exciting.”

The predictions, based on which Stewart organized the trip, were made by a machine-learning algorithm. Using satellite data from locations known to have high density of the fungi population, Stewart and his colleagues trained models to predict regions around the world that are biodiversity hotspots for mycorrhizal fungi. Then, the team would make their way to those locations to collect samples for DNA analysis to corroborate the model’s findings.

The Society for the Protection of Underground Networks (SPUN), where Stewart is an ecological data scientist, is an initiative that’s working to map the presence of mycorrhizal fungi around the world, while also attempting to identify underground ecosystems where the species are most at threat. Since 2021, the group of scientists has been working with the goal of spreading awareness about the fungi, identifying their locations, and using the data to advocate for their protection.

While fungi conservation might not seem like a crucial mission, its impacts are critical for the health of the planet.

“When most people think of fungi, they think of mold on bread,” Stewart said. “But there’s more to it than that.”

Mycorrhizal fungi are found in almost every soil in the world. For about 400 million years, these organisms have formed important symbiotic relationships with plants. It’s an ancient association that’s been beneficial for the plants as well as the fungi that live on their roots. The fungi help plants extract nutrients and water from the soil, and keep root pathogens at bay. In return, during photosynthesis, the plants feed carbon to the fungal networks. As a result, these fungi act as repositories of carbon, making them an important tool for carbon sequestration, and subsequently in efforts to mitigate the effects of climate change.

According to a study published recently in the journal Current Biology, more than 13 billion metric tons of carbon dioxide moves from plants to mycorrhizal fungal networks every year. That’s the equivalent of about 36% of annual fossil fuel emissions. However, several factors — including agricultural expansion, deforestation, and increased use of chemical fertilizers — pose a threat to the ability of these species to do their job.

“They are critical players in global ecosystems for cycling not just carbon, but also phosphorus and other nutrients,” Stewart said. “But despite that, we are not thinking about what is happening below ground to these organisms that are so integral to the global structure of vegetation and the flows of nutrients.”

Despite their vital role, these fungi have not been studied extensively, largely due to technological and logistical hurdles. That’s the gap SPUN is attempting to fill.

With its work to map the fungi and identify biodiversity hotspots, SPUN is attempting to incorporate the protection of these fungi into conservation and policy agendas. Stewart said it’s the rapid growth of technological tools that’s made it possible to do this work.

“This work wouldn’t have happened five or 10 years ago because we didn’t have the technology to do what we are doing now with confidence and at scale,” he said. “Earlier, we couldn’t determine the presence of the fungi unless we dug up a root, stained it, and did all these horrible things with acid to it to look under a microscope.”

Recent improvements in collecting and analyzing environmental DNA, or eDNA, have made the work much easier. Analyzing eDNA involves collecting samples and determining the species present in a location based on any genetic material they may have left behind. “That really propelled and exploded the way we were able to talk about the diversity of these organisms,” Stewart said.

However, given the ubiquitous nature of mycorrhizal fungi, determining the areas to conduct eDNA analysis isn’t always easy. That’s where remote-sensing data and machine-learning technologies step in.

The team at SPUN use data from GlobalFungi, a database that maps the occurrences of the organisms around the world, and layers it with remote-sensing data — including data on vegetation types, temperature, and relationship between the fungi, the environment and the climate they live in — to train machine-learning models in collaboration with the Crowther Lab at ETH Zürich.

“We then task the machine-learning algorithm to extrapolate and give us the best prediction outside of anywhere we have ever been to,” Stewart said. “It tells us where to go and gives us a map of the most underexplored and unknown places for fungal diversity on Earth.”

Once the machine-learning algorithms identify the global hotspots, team members at SPUN conduct field trips to collect samples and corroborate the information using DNA testing and analysis. Based on the predictions made by the algorithm, the team has so far successfully conducted field trips to places including Kazakhstan, the Galápagos Islands in Ecuador, and the Patagonia region shared by Argentina and Chile.

However, Stewart said limitations still exist.

For one, he said, higher-resolution satellite imagery would make their work more effective. “I want to know what’s happening on every square centimeter,” he said.

There also needs to be more data collection to better train the machine-learning models, which by definition are limited by the amount of data they’re fed.

“If we can only describe 70% of the Earth’s environmental heterogeneity, that is 70% of the different ways climate interacts on Earth, we are limited to only really predicting within that range of data,” Stewart said.

In the coming years, he added, SPUN aims to work through these hurdles to finish the first version of the mycorrhizal fungi maps and publish them as open-source data for public use so that “we can potentially start coming up with preventative solutions for biodiversity loss.”

The group is also working to provide financial and technical support to researchers, scientists and local communities in countries including Ghana, Mongolia and Mexico to map mycorrhizal fungi in their home ecosystems.

“We are not trying to do it all,” Stewart said. “We also really want to help to ‘de-Northern Hemispherize’ science and bring it to places where they might not have the same funding or access to technology.”

Banner image: Greville’s bolete, which has a mycorrhizal relationship with larch trees. Image by Ludo Dolu via Flickr (CC BY-NC-SA 2.0).

Abhishyant Kidangoor is a staff writer at Mongabay. Find him on Twitter @AbhishyantPK.

Citation:

Hawkins, H.-J, Cargill, R. I. M., Van Nuland, M. E., Hagen, S. C., Field, K. J., Sheldrake, M., … Kiers, E. T. (2023). Mycorrhizal mycelium as a global carbon pool. Current Biology, 33(11), R560-R573. doi:10.1016/j.cub.2023.02.027