EcoFABs Make Plant Microbiome Experiments Reproducible-Across Continents and Ready for AI

EcoFABs make plant-microbiome experiments reproducible across continents - and ready for AI

AI can help predict how microbial communities affect crops and soils. But AI only learns from large, consistent datasets - and those are rare in plant-microbiome research.

That gap just narrowed. In a first-of-its-kind global study published in PLOS Biology, an international team used standardized EcoFAB growth chambers to run identical plant-microbe experiments across labs on three continents and got matching results.

Key takeaways

• EcoFABs are takeout-box-sized chambers that let scientists grow plants with defined microbiomes under tightly controlled conditions.
• Identical EcoFAB experiments across multiple labs produced consistent results, supported by shared protocols, tools, and datasets.
• These reliable, large-scale datasets are well-suited for training AI models to advance crop development, soil health, and agriculture.

Reproducibility has been the sticking point

Plant microbiomes influence nutrient cycling, disease resistance, and growth. Yet comparing results across labs has been difficult due to variation in methods, materials, and environments.

EcoFABs change that. These clear plastic chambers, developed at Berkeley Lab, standardize how plants and selected microbes are co-cultured and monitored. As Vlastimil Novak put it, "If you want to make meaningful predictions about microbes and plants, especially with future AI models, you need clean, consistent datasets. EcoFABs provide exactly that."

The team also leveraged years of investment in model grasses and genomic resources, creating a system where the microbiome, environment, and plant genetics could be controlled and compared.

A worldwide test - same kits, same steps, same outcomes

Labs in the United States, Germany, and Australia received EcoFAB kits with the same seeds, a common set of 16 or 17 microbes, and step-by-step protocols. Shipping live microbes was a challenge - strict safety rules and lots of dry ice - but every site ran the experiment successfully.

The only difference between treatments was one strain: Paraburkholderia sp. OAS925. When included, it consistently dominated the root environment and slightly reduced plant size. That pattern held across all participating labs.

The team also measured root exudates - small molecules that shape plant-microbe interactions. Across dozens of compounds, profiles largely matched between labs. A few unstable molecules (for example, dopamine) varied more, and small temperature differences affected plant size, but the overall signal remained consistent.

"The unique design of the EcoFAB allows us to monitor multiple interaction partners with different levels of complexity, reproducibly across labs worldwide," said Borjana Arsova. "Because we can precisely control growth conditions and track development over time, we can start to untangle these multidirectional relationships - and ultimately apply that knowledge to improve agriculture."

Why this matters for AI and agriculture

Training useful models requires standardized inputs. EcoFABs generate comparable, high-quality datasets on plant-microbe dynamics - exactly what modelers need to move from description to prediction.

The study also quantified inter-lab variation, giving AI practitioners the context needed to integrate data from multiple sources. As Trent Northen noted, those baselines are essential for building models that generalize beyond a single lab bench.

Michelle Watt summed up the practical impact: translation to field settings has lagged because results vary too much between labs. The global EcoFAB study shows that barrier can be overcome.

Tools you can use now

• Access EcoFAB 2.0 devices through JGI's Community Science Program and FICUS calls (proposal-based, DOE relevance).
• Use the same 17-member synthetic community via DSMZ to replicate experiments or build new ones.
• Follow open, step-by-step protocols (with narrated videos) to reduce variability and improve comparability.
• Benchmark against the study's openly available datasets via the National Microbiome Data Collaborative.

What the data say - and what's next

EcoFABs produced consistent outcomes across continents: a known root colonizer reliably dominated when present, plant growth responses were repeatable, and most metabolite patterns aligned across labs. That's the foundation needed to scale experiments and train predictive models.

Next up: automation. The team is pairing EcoFABs with robotics and sensors (the "EcoBOT") to run experiments automatically and stream high-quality data for analysis at scale.

"Rigorous science means getting consistent results when asking the same question," said Paul Schulze-Lefert. "By collecting data from laboratories on three continents, this work shows reproducibility is possible in plant microbiome research - an essential step toward developing microbial probiotics for real-world agriculture."

Collaboration, access, and licensing

EcoFAB devices and associated resources are available to the community through DOE user facilities and partner programs. Companies interested in licensing EcoFAB technology for production can contact Berkeley Lab's Intellectual Property Office at ipo@lbl.gov.

This work was supported in part by the U.S. Department of Energy's Office of Science. For information on DOE Office of Science programs, visit energy.gov/science.