Flavor-cyber-agriculture: Optimization of plant metabolites in an open-source control environment through surrogate modeling

Arielle J. Johnson, Elliot Meyerson, John de la Parra, Timothy Savas, Risto Miikkulainen, Caleb HarperPLoS 2019 · DOI · OpenAlex

Abstract

Food production in conventional agriculture faces numerous challenges such as reducing waste, meeting demand, maintaining flavor, and providing nutrition. Contained environments under artificial climate control, or cyber-agriculture, could in principle be used to meet many of these challenges. This paper describes a method for optimizing flavor by combining cyber-agriculture, metabolomic phenotype (chemotype) measurements, and machine learning. In a pilot experiment, environmental conditions (photoperiod and UV light) were applied to basil plants inside a hydroponic farm. Flavor-active volatile molecules were measured using gas chromatography-mass spectrometry (GC-MS), and symbolic regression was used to construct a surrogate model. This model discovered that a 24-hour photoperiod induces the most flavor molecule production, effectively doubling the volatile profile compared to control groups.

Introduction: The Dilution Effect in Modern Agriculture

For decades, conventional agriculture has prioritized biomass and yield, often at the expense of nutritional density and flavor—a phenomenon known as the "dilution effect." As crops are optimized for size, growth speed, and shelf-stability, the concentration of secondary metabolites, which define flavor and health benefits, has declined. Essentially, modern farming has traded quality for quantity.

Optimizing for flavor is notoriously complex. "Flavor" is not the product of a single gene but a complex expression of dozens of volatile molecules influenced by a vast array of environmental variables, including light spectrum, water availability, nutrient density, and temperature. Traditional botanical experiments, which often change one variable at a time, are too slow and expensive to navigate this high-dimensional search space. This study proposes a "cyber-agriculture" approach, leveraging automated controlled environments and machine learning to rapidly discover optimal "climate recipes."

Methodology: Cyber-Agriculture and Symbolic Regression

The researchers utilized an open-source hydroponic system called the Food Server, a multi-tray version of the MIT "Food Computer." This system allows for precise control over the plant's environment. The study focused on three primary environmental variables: photoperiod (hours of light per day), UV period (hours of supplemental UV-B), and Photosynthetic Active Radiation (PAR) intensity.

The experimental workflow followed a three-round iterative process:

  1. Measurement: Sweet Basil (Ocimum basilicum) was grown under various conditions. After harvest, 17 key volatile molecules were measured via Gas Chromatography-Mass Spectrometry (GC-MS).
  2. Metric Creation: To simplify the complex chemical data, the team developed a "chemscore" and "R-Score." These metrics aggregated the 17 chemical concentrations into a single value representing flavor density relative to a control group.
  3. Surrogate Modeling: Instead of using "black box" neural networks, the researchers employed Symbolic Regression. This technique uses evolutionary algorithms to find the simplest mathematical equations that describe the relationship between inputs (light) and outputs (flavor). This approach provides interpretable models that reveal the underlying biological drivers.
  4. Optimization: The surrogate model was used to predict "recipes" that would maximize flavor, which were then implemented and tested in subsequent growing rounds.

Key Insights: The 24-Hour Photoperiod

The most significant discovery was the efficacy of a 24-hour photoperiod. While standard botanical wisdom suggests plants require a period of darkness for optimal health and circadian rhythm, the results indicated that constant light maximized the production of flavor-active volatiles in basil. This condition, impossible in nature, represents a unique advantage of cyber-agriculture.

Furthermore, symbolic regression revealed that light duration was the "master switch" for basil flavor. The most parsimonious model was often a simple linear relationship with photoperiod, suggesting that duration far outweighed UV or PAR intensity in this specific setup. The study also mathematically confirmed the "dilution effect," showing a clear negative correlation between plant weight and flavor density—the "best" tasting plants were often smaller.

Results and Performance

In the third experimental round, ML-optimized recipes achieved a volatile profile (R-Score) double that of the control group. The linear model trained on all variables achieved an $R^2$ of 0.577. Interestingly, a model using only photoperiod achieved an $R^2$ of 0.550, proving that light duration was the dominant driver. The surrogate model successfully extrapolated into unexplored regions of the search space, correctly predicting that moving toward 24-hour light cycles would yield superior results.

Conclusion and Future Directions

This research demonstrates a successful proof-of-concept for using cyber-agriculture and machine learning to reverse the dilution effect. By treating the plant's environment as a programmable interface, researchers can discover "climate recipes" that would be difficult to find through traditional means.

Future work will expand the search space to include temperature, humidity, and nutrient density. The researchers also aim to move from static recipes to time-varying recipes, applying specific stressors (like high UV or drought) at strategic growth stages to further enhance plant phenotypes. This methodology holds potential for optimizing not just flavor, but also fiber strength in cotton or the concentration of medicinal compounds in pharmaceutical plants.

Summary updated

Abstract

Food production in conventional agriculture faces numerous challenges such as reducing waste, meeting demand, maintaining flavor, and providing nutrition. Contained environments under artificial climate control, or cyber-agriculture, could in principle be used to meet many of these challenges. Through such environments, phenotypic expression of the plant-mass, edible yield, flavor, and nutrients-can be actuated through a "climate recipe," where light, water, nutrients, temperature, and other climate and ecological variables are optimized to achieve a desired result. This paper describes a method for doing this optimization for the desired result of flavor by combining cyber-agriculture, metabolomic phenotype (chemotype) measurements, and machine learning. In a pilot experiment, (1) environmental conditions, i.e. photoperiod and ultraviolet (UV) light (known to affect production of flavor-active molecules in edible plants) were applied under different regimes to basil plants (Ocimum basilicum) growing inside a hydroponic farm with an open-source design; (2) flavor-active volatile molecules were measured in each plant using gas chromatography-mass spectrometry (GC-MS); and (3) symbolic regression was used to construct a surrogate model of this chemistry from the input environmental variables, and this model was used to discover new combinations of photoperiod and UV light to increase this chemistry. These new combinations, or climate recipes, were then implemented in the hydroponic farm, and several of them resulted in a marked increase in volatiles over control. The process also led to two important insights: it demonstrated a "dilution effect", i.e. a negative correlation between weight and desirable chemical species, and it discovered the surprising effect that a 24-hour photoperiod of photosynthetic-active radiation, the equivalent of all-day light, induces the most flavor molecule production in basil. In this manner, surrogate optimization through machine learning can be used to discover effective recipes for cyber-agriculture that would be difficult and time-consuming to find using hand-designed experiments.
BibTeX
@inproceedings{johnson2019flavor,
  author = {Arielle J. Johnson and Elliot Meyerson and John de la Parra and Timothy Savas and Risto Miikkulainen and Caleb Harper},
  title = {{Flavor-cyber-agriculture: Optimization of plant metabolites in an open-source control environment through surrogate modeling}},
  booktitle = {PLoS},
  year = {2019},
  doi = {10.1371/journal.pone.0213918},
}

Loading related content...

Figure