Machine Learning Designs Microbial Communities That Boost Tomato Growth and Heat Tolerance

Tomatoes, one of the most widely cultivated vegetable crops in the world, may soon owe part of their resilience to an unlikely source: carefully engineered communities of soil bacteria designed not by trial and error in the greenhouse, but by a machine-learning algorithm. Researchers at Kyoto University, working with colleagues at Tohoku University and RIKEN, have demonstrated that a data-driven approach to building defined microbial communities can promote tomato growth and strengthen the plants’ tolerance to high-temperature stress. The study, published in The ISME Journal, offers a glimpse of how synthetic ecology and artificial intelligence could converge to address one of agriculture’s most pressing challenges: sustaining crop yields on a warming, increasingly drought-prone planet.

Plants are never truly alone. Their roots, leaves, and surrounding soil harbor a dense and diverse array of microorganisms that influence nearly every aspect of plant life, from nutrient uptake to immune signaling. These microbial partners help plants absorb minerals, fend off pathogens, and adapt to environmental stresses. As human-induced climate change intensifies, bringing with it more frequent heat waves and prolonged droughts, crop yields have suffered, and researchers have increasingly turned to biofertilizers — preparations of beneficial rhizosphere microorganisms — as a biological alternative or supplement to chemical inputs. The idea is elegant in principle: harness the microbes that plants already cooperate with in nature and deploy them deliberately in the field.

In practice, however, single-strain biofertilizers have proven fragile. A lone microorganism introduced into a crop’s root zone must survive in an environment that fluctuates dramatically in temperature, moisture, nutrient availability, and competition from resident microbes. Many inoculants fail to establish themselves, and even those that persist often lose their functional benefits over time. This fragility has pushed scientists toward a more sophisticated strategy: defined microbial communities, or DMCs, in which multiple microorganisms are deliberately combined. Within such communities, members influence one another’s survival and activity, exchanging metabolites and creating ecological niches that can make the whole assemblage more stable and resilient than any of its parts.

Yet designing an effective DMC is far from straightforward. The composition and function of a microbial community shift depending on the plant-specialized metabolites secreted by the host and on prevailing environmental conditions. A community that thrives under laboratory conditions may collapse in the field, or a combination that benefits one crop variety may do nothing for another. The number of possible combinations of bacterial strains and metabolites is astronomically large, making conventional trial-and-error experimentation impractical. It was precisely this combinatorial explosion that motivated the Kyoto University team to seek a rational, predictable method for community design — one grounded in data rather than intuition.

Co-corresponding author Akifumi Sugiyama of Kyoto University explained the conceptual shift behind the work. “I became particularly interested in moving beyond simple one-to-one relationships to understand how these metabolite-mediated interactions unfold within the diverse microbial communities found in soil, and how they influence plant growth,” he said. Rather than studying individual microbe-plant pairs in isolation, the team set out to capture the web of interactions that emerges when multiple bacteria and plant metabolites coexist — the kind of complexity that determines whether a microbial community actually helps a crop or quietly fades away.

The experimental design was methodical. The researchers first isolated bacteria from the roots of tomato plants, ensuring that the strains they worked with were natural associates of their target crop. They then combined nine types of bacteria and two types of plant metabolites in various patterns to create a panel of defined microbial communities. These communities were inoculated onto tomato roots, and the plants were cultivated at varying temperatures. By systematically varying both the community composition and the growth conditions, the team generated a rich dataset linking microbial assemblages, metabolites, temperature regimes, and plant performance.

The analytical engine of the study was a machine-learning model built with the elastic net regression algorithm, a technique well suited to datasets in which many predictor variables are correlated with one another — a hallmark of ecological data. Trained on the experimental results, the model learned to predict the aboveground fresh weight of tomato plants, a key indicator of growth, from the composition of the microbial community and the environmental conditions. Once validated, the model could be run in reverse: instead of predicting the outcome of a given community, the researchers could ask the algorithm to design a new community predicted to have a strongly positive effect on tomato growth.

The algorithm’s recommendation proved its worth. A defined microbial community designated DMC G2, composed of six bacterial strains together with the metabolite tomatine, promoted tomato growth in laboratory tests and, crucially, improved the plants’ tolerance to high-temperature stress. The team then took the experiment outdoors, cultivating tomatoes in field-like conditions to see whether the benefits would survive contact with the messiness of the real world. They did: outdoor trials showed that DMC G2 significantly increased the aboveground fresh weight of the tomatoes, and gene expression analysis revealed molecular signatures of enhanced high-temperature tolerance in the treated plants. The combination of growth promotion and stress tolerance in a single, rationally designed community is what makes the result notable.

Achieving this level of rigor demanded extraordinary experimental discipline. “We needed to cultivate tomatoes under a wide variety of conditions with high precision, so I would like to thank the team members who carried out these repetitive yet exacting experiments over a period of approximately two years,” Sugiyama said. The two-year effort underscores a central reality of this emerging field: machine learning can dramatically narrow the search space of possible microbial combinations, but generating the high-quality training data that such models require still depends on meticulous, labor-intensive plant cultivation and measurement under controlled and varied conditions.

The broader implications extend well beyond tomatoes. The study demonstrates that data-driven microbial community design can improve crop stress tolerance, and the authors emphasize that their versatile methods are expected to extend broadly to other crops, holding significant potential for enabling sustainable agriculture. As climate change continues to erode yields and conventional agricultural inputs face mounting environmental scrutiny, the ability to rationally engineer plant-associated microbiomes — predicting, rather than guessing, which combinations of bacteria and metabolites will help a crop thrive — could become a cornerstone of resilient farming systems. The tomato, in this telling, is not just a crop but a proof of concept: evidence that the invisible ecosystems around plant roots can be understood, modeled, and ultimately designed to humanity’s benefit.

Subject of Research: Machine learning-guided design of defined microbial communities to enhance tomato plant growth and heat-stress tolerance

Article Title: Optimizing tomato plant growth with microbes

Article References: Optimizing tomato plant growth with microbes. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: tomato, microbial communities, machine learning, biofertilizers, plant growth, heat stress tolerance, rhizosphere, elastic net regression, tomatine, sustainable agriculture, Kyoto University, synthetic ecology