Abstract :
[en] Capparis spinosa L. is a Mediterranean medicinal species of high economic value, yet its large-scale propagation and metabolite production remain constrained by conventional approaches. This study evaluated the effects of four hormonal combinations, 6-benzylaminopurine (BAP) + 1-naphthaleneacetic acid (NAA), BAP + 2,4-dichlorophenoxyacetic acid (2,4-D), kinetin (KIN) + 2,4-D, and KIN + NAA, on callus fresh weight gain (FWG) from leaf explants using a pairwise combinatorial experimental design. The resulting dataset was analyzed using three machine learning (ML) algorithms, Random Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGBoost), as well as a second-degree polynomial regression model (PL2). Among these models, RF achieved the best performance, with a cross-validation coefficient of determination (R²CV) of 0.89, a coefficient of determination (R²) of 0.86, and a root mean square error (RMSE) of 263.33. SHapley Additive exPlanations (SHAP) analysis revealed 2,4-D as the most influential predictor of callogenesis, followed by a positive contribution from BAP, while KIN and NAA had minimal or negative effects on FWG prediction. Experimental validation was restricted to the five highest-ranked RF-predicted combinations, demonstrating good agreement between predicted and observed values. In addition, rutin, the major bioactive flavonoid of C. spinosa, was identified by LC-QTOF-MS/MS and relatively quantified by LC-TQ-MS/MS under BAP and 2,4-D combinations. A stacked RF model integrating FWG predictions was further developed to estimate rutin accumulation, achieving satisfactory performance with an R² of 0.82, an R²CV of 0.84, and an RMSE of 0.41. Maximum accumulation was observed at moderate hormone concentrations. Overall, this integrative ML and SHAP-based approach provides an interpretable and scalable strategy for optimizing callus culture and enhancing metabolite production, offering a sustainable alternative to wild plant harvesting.
