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

摘要: 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 principle be used to meet many of these challenges. Through environments, phenotypic expression the plant—mass, edible yield, nutrients—can actuated through a “climate recipe,” where light, water, nutrients, temperature, other ecological variables are optimized achieve desired result. This paper describes method for doing this optimization result flavor by combining metabolomic phenotype (chemotype) measurements, machine learning. In pilot experiment, (1) environmental conditions, i.e. photoperiod ultraviolet (UV) light (known affect flavor-active molecules plants) were applied different regimes basil plants (Ocimum basilicum) growing inside hydroponic farm with an open-source design; (2) volatile measured each plant using gas chromatography-mass spectrometry (GC-MS); (3) symbolic regression was construct surrogate model chemistry from input variables, discover new combinations UV increase chemistry. These combinations, recipes, then implemented farm, several them resulted marked volatiles over control. The process also led two important insights: it demonstrated “dilution effect”, negative correlation between weight desirable chemical species, discovered surprising effect that 24-hour photosynthetic-active radiation, equivalent all-day induces most molecule basil. manner, learning can effective recipes cyber-agriculture would difficult time-consuming find hand-designed experiments.