Researchers have used artificial intelligence to catalog thousands of symbiotic microbes that live in hidden partnerships with other organisms, a discovery that could reshape biotechnology and our understanding of microbial life. An international team led by the U.S. Department of Energy's Joint Genome Institute published findings on August 31, 2026, in Nature Biotechnology showing that 15-23% of uncultivated microbes across ecosystems may actually be living in these intimate relationships with host organisms, far more than previously appreciated.

For decades, microbial science has faced a fundamental blind spot. Most bacteria and archaea that live in partnership with other organisms cannot be grown in laboratory conditions, so their genomes have been systematically underrepresented in scientific databases. This meant researchers could catalog only a fraction of microbial diversity, missing entire ecosystems of specialized microbes that drive the health and functioning of natural systems. The new AI tool, called symclatron, scans hundreds of thousands of microbial genomes to identify which ones live in partnership with a host, bringing this hidden world into view.

What Makes These Symbiotic Microbes So Hard to Study?

The challenge has always been practical. Traditional microbiology relies on isolating microbes from environmental samples and growing them in the lab, a technique that works for only a small fraction of microbial life. The vast majority of microbes cannot be cultivated this way, and for years, scientists didn't fully understand why. The new research suggests a key reason: many of these uncultivated microbes are actually dependent on living inside or on another organism to survive. They have evolved specialized metabolic capabilities through millions of years of co-evolution with their hosts, making them impossible to grow alone.

The research team analyzed more than 31,000 publicly available environmental sequencing projects from the Joint Genome Institute's Integrated Microbial Genomes and Microbiomes database, one of the world's largest repositories of microbial genomic data spanning soils, deep subsurface habitats, host-associated microbiomes, marine and freshwater systems, and more across all major continents and oceans. Using AI, they identified more than 14,000 genomes belonging to symbionts, microbes that live in close, sustained association with a host.

"This work moves us toward a more predictive field by providing a robust framework to identify host-dependence directly from sequence data," said Frederik Schulz, co-corresponding author of the study. "This allows us to move beyond simply cataloging diversity to understanding the ecological roles of uncultivated microbes at scale."

Frederik Schulz, Co-corresponding Author, Joint Genome Institute

How Did the AI Model Learn to Spot Symbionts?

The symclatron AI tool works by analyzing the functional gene content of microbial genomes rather than relying on taxonomy alone. The model achieved 96% overall accuracy when tested on entire branches of the microbial tree it had never encountered before, a critical validation that shows it can generalize to novel microbial lineages. This is a meaningful technical achievement because microbial lifestyles have historically been nearly impossible to infer from genome sequence data alone, especially for microbes that have never been grown in a laboratory.

The research team took an important approach to building the model: they relied on manual expert curation of primary literature rather than automated database annotations. This careful validation strategy prevents the propagation of erroneous or circular metadata that could undermine the model's reliability. The symclatron tool and the resulting Symbiont Genomes catalog, or SymGs, are now publicly available resources that span roughly half of all known bacterial and archaeal phyla.

What Do These Symbiotic Relationships Reveal About Microbial Evolution?

The findings reveal that symbiotic lifestyles reshape microbial metabolism in two distinct directions. Many symbionts have shed parts of the metabolism required for independent life, particularly the ability to produce their own amino acids and DNA components, relying instead on their hosts for these essential molecules. But the researchers also found evidence that some symbionts encode for metabolic capabilities not present in closely related free-living bacteria, including more complete pathways for vitamin biosynthesis and transporting molecules across membranes.

These specialized metabolic adaptations are the result of prolonged co-evolution with hosts. Some symbionts have adapted to having smaller, simpler genomes through their partnership, which could inform the biodesign of industrially relevant microbes with novel pathways and enzymatic tools. By cataloging which microbes carry which metabolic tools, SymGs offers a genomic resource that could inform future research into microbial partnerships, including potential applications in biotechnology, bioenergy, biomanufacturing, and bioproducts.

How Can Scientists Use This Discovery for Biotechnology?

• Gene Discovery: The SymGs catalog provides access to uncultivated microbial lineages that cannot be analyzed through traditional cultivation, opening new sources of genes and metabolic pathways for industrial applications.
• Metabolic Engineering: Understanding which genes symbionts have lost and which they have gained allows researchers to design microbial systems with novel capabilities for energy production and biomanufacturing.
• Ecosystem Understanding: The AI-driven identification of symbiotic relationships helps researchers understand the ecological roles of uncultivated microbes at scale, revealing how these hidden partnerships drive ecosystem health and functioning.

"This approach identifies symbiotic lifestyles based on functional gene content rather than taxonomy alone," explained Juan C. Villada, a data scientist with the Joint Genome Institute and one of the study's lead authors. "The data allows access to uncultivated microbial lineages that cannot be analyzed through traditional cultivation."

Juan C. Villada, Data Scientist, Joint Genome Institute

The scale of this work was made possible by the SymGs data consortium, composed of researchers from nine institutions who had previously worked with the Joint Genome Institute through various proposals. This collaboration brought together vast collections of metagenomic data from across Earth's biomes, with publicly available datasets serving as the raw material for discovery. Researchers from Bigelow Laboratory for Ocean Sciences, Stanford University, UNSW Sydney, University of Wisconsin-Madison, University of British Columbia, University of Illinois Urbana-Champaign, University of East Anglia, Michigan State University, and Ohio State University all contributed to the study.

The implications of this research extend far beyond basic science. By providing a computational bridge to identify microbial lineages that are likely difficult or impossible to grow in isolation, symclatron offers a high-confidence tool for prioritizing which microbes to study further. While traditional culturing remains the gold standard for validation, this AI-driven approach dramatically expands the frontier of microbial biology, revealing that symbiosis is likely a key reason why most microbes resist laboratory cultivation. As researchers continue to explore this hidden world of microbial partnerships, new opportunities for biotechnology and disease treatment may emerge from understanding how these ancient relationships shape life on Earth.