Novel application of machine learning models to predict secondary metabolites in Dracocephalum moldavica L

This study aimed to predict the levels of key secondary metabolites-carotenoids, the amino acid proline, and soluble sugars-in Dracocephalum moldavica under salinity stress using artificial neural networks (ANNs). Two ANN models, namely the Radial Basis Function (RBF) and Multilayer Perceptron (MLP), were employed.The models used 13 morphological and physiological traits, including leaf length, leaf width, number of leaves, number of internodes, crown diameter, stem length, internode length, number of lateral stems, lateral stem length, root length, shoot fresh weight, shoot dry weight, and relative water content, as inputs. Model performance was evaluated using a 5-fold cross-validation approach to ensure robust and reliable predictions.The RBF model consistently outperformed the MLP model, demonstrating a superior ability to capture nonlinear relationships between morphological traits and biochemical responses. While the MLP model provided reasonable estimates, it was less accurate for all target metabolites.Within the scope of the current dataset and greenhouse conditions, the RBF model proved to be a reliable tool for predicting secondary metabolites. This approach offers a rapid and cost-effective method to estimate key biochemical compounds, although predictions should be interpreted with caution. Future studies using independent external datasets are encouraged to further assess the model’s generalizability.