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

Abstract

30Food production in conventional agriculture faces numerous challenges such as reducing 31 waste, meeting demand, maintaining flavor, and providing nutrition. Contained environments 32 under artificial climate control, or cyber-agriculture, could in principle be used to meet many of 33 these challenges. Through such environments, phenotypic expression of the plant---mass, edible 34 yield, flavor, and nutrients---can be actuated through a "climate recipe," where light, water, 35 nutrients, temperature, and other climate and ecological variables are optimized to achieve a 36 desired result. This paper describes a method for doing this optimization for the desired result of 37 flavor by combining cyber-agriculture, metabolomic phenotype (chemotype) measurements, and 38 machine learning. In a pilot experiment, (1) environmental conditions, i.e. photoperiod and 39 ultraviolet (UV) light (known to affect production of flavor-active molecules in edible plants) 40 were applied under different regimes to basil plants (Ocimum basilicum) growing inside a 41 hydroponic farm with an open-source design; (2) flavor-active volatile molecules were 42 measured in each plant using gas chromatography-mass spectrometry (GC-MS); and (3) 43 symbolic regression was used to construct a surrogate model of this chemistry from the input 44 environmental variables, and this model was used to discover new combinations of photoperiod 45 and UV light to increase this chemistry. These new combinations, or climate recipes, were then 46 implemented in the hydroponic farm, and several of them resulted in a marked increase in 47 volatiles over control. The process also led to two important insights: it demonstrated a "dilution 48 effect", i.e. a negative correlation between weight and desirable chemical species, and it 49 discovered the surprising effect that a 24-hour photoperiod of photosynthetic-active radiation, the 50 equivalent of all-day light, induces the most flavor molecule production in basil. In this manner, 51 surrogate optimization through machine learning can be used to discover effective recipes for 52 3 cyber-agriculture that would be difficult and time-consuming to find using hand-designed 53 experiments. 54 55 1 Introduction 56 The so-called "dilution effect," noted since the 1940's and systematically reviewed since 57 the early 1980's [1], describes an inverse relationship between yield and nutrient concentration in 58 food: For many nutritionally-important chemical constituents of food plants, such as minerals, 59 protein, and vitamins, an increase in biomass is accompanied by a decrease in nutrient 60 concentration. This effect has been systematically demonstrated in historical nutrient content 61 studies over the last 50-70 years [2,3], as well as in controlled side-by-side trials that have shown 62 a relationship between nutrient dilution and genetics [4], artificial fertilization [5], and elevated 63 carbon dioxide levels related to climate change [6,7]. Flavor, known to be an important element 64 of food and of ea...

Actionable Subgroup Discovery and Urban Farm Optimization

“…Interestingly, the RMSE seems to be quite good at first glance, but this does not guarantee that the model will behave correctly on all elements of the search space: on the best recipe found, it largely overestimates the yield, leading to a non-interesting recipe. It seems that this method performs poorly on recipes with more attributes than in [9]. Further studies are here needed.…”

“…Let us now compare our framework to the Surrogate Modeling method presented in [9]. To be fair, we give the same number of data points to build the Symbolic Regression surrogate model as we used in previous experiments, i.e., 150 for training the model (we evaluated the RMSE of the model on a test set of 38 other samples).…”

“…To be fair, we give the same number of data points to build the Symbolic Regression surrogate model as we used in previous experiments, i.e., 150 for training the model (we evaluated the RMSE of the model on a test set of 38 other samples). We use gplearn [19], with default parameters, except for the number of generations and the number of models evaluated for each generations, which are respectively of 1000 and 2000, as in [9]. Note that the model obtained has a RMSE of 2112, and it is composed of more than 2000 terms (including mathematical operators), therefore the argument of interpretability is questionable.…”

“…For model-based methods, the idea is to build a model simulating the ground truth using available data and then to use it to guide the search process. For instance, [9] introduced a solution for recipe optimization using this type of method with the goal of optimizing the flavor of plants. Their framework is based on using a surrogate model, in this case a Symbolic Regression [13].…”

“…It enables the growth of plants in fully controlled environments close to the place where consumers are [8]. Most of the crop protection chemical products can be removed while being able to optimize both the quantity and the quality of plants (e.g., improving the flavor [9] or their chemical proportions [20]). Urban farms can generate large amounts of data that can be pushed towards a cloud environment such that various machine learning and data mining methods can be used.…”