Predictive Chromatography of Leaf Extracts Through Encoded Environmental Forcing on Phytochemical Synthesis

Environment fluctuations can influence a plant's phytochemical profile via phenotypic plasticity. This adaptive response ensures a plant's survival under fluctuating growth conditions. However, the resulting plant extract composition becomes unpredictable, which is a problem for highly standardized medicinal applications. Here we demonstrate, for the first time, the feasibility of tracking the changes in the phytochemical profile based on real-time measurements of a few environment and extract-preparation variables. As a result, we predicted the chromatograms of Blumea balsamifera extracts through an imputation-augmented convolutional neural network, which uses the image-transformed temporal measurements of the variables. We developed a sensor network that collected data in a greenhouse and a training algorithm that concurrently generated a data representation of the implicit plant-environment interactions leading to the mutable chromatograms of leaf extracts. We anticipate the generic applicability of the method for any plant and recognize its potential for addressing the standardization problems in plant therapeutics.