Predicting Camellia genus flower traits using leaf metabolomics and machine learning approaches

Artificial intelligence (AI), and in particular machine learning (ML), is revolutionising plant science by enabling predictive models to be created from complex omic datasets, notably metabolomics. For slow-growing species, such as Camellia , an emblematic genus in cosmetics, horticulture, and food industry, ML offers a unique opportunity to accelerate varietal selection and screening for bioactive potential without waiting for phenotypic maturity [ 1 ]. This study explores the predictive potential of leaf metabolomics for anticipating floral characteristics and biological activities of interest. More than 2,000 samples, representing more than 300 genotypes and around twenty species of the Camellia genus, were analysed using reverse-phase liquid chromatography coupled with Orbitrap high-resolution mass spectrometry (LC-MS, negative ion mode). The metabolomic profiles obtained from the leaves were used to train models capable of predicting flower shape and colour with an accuracy of over 80%, this model could be applied to young plants by predicting phenotype traits up to 3-5 years before flowering. Predictive models were validated using different independent sampling years. The metabolic predictor identified include more than a dozen chemical families, dominated by phenolic compounds, lipids, amino acid derivatives, and benzoin-type compounds. This approach and the richness of the dataset acquired can also be used to predict new parameters related to the plant, its crop or the biological activities that could be associated with it. Publication History Article published online: 05 March 2026 © 2026. Thieme. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany