In vitro seed germination of cannabis as the first physiological stage in the plant life cycle is not only important for studying factors affecting cultivation conditions but also crucial for obtaining juvenile tissue as a potential explant for different in vitro procedures. On the other hand, in vitro seed germination is a multi-variable biological process that can be influenced by genetic (genotype) and physical factors (medium composition and environmental conditions). Therefore, a powerful mathematical methodology such as artificial neural networks (ANNs) is well suited to analyze the data and optimize the conditions this complex system. The current study was aimed to evaluate the effect of different types and concentrations of carbohydrate sources (sucrose and glucose) as well as different strengths of DKW (Driver and Kuniyaki Walnut) and mMS (Murashige and Skoog Medium, Van der Salm modification) media on seed germination indices as well as morphological features of in vitro-grown cannabis seedlings by using three ANNs including multilayer perceptron (MLP), radial basis function (RBF), and generalized regression neural network (GRNN). The GRNN model displayed higher predictive accuracy (r²>0.70) in both training and testing sets for all germination indices and morphological traits in comparison to RBF or MLP. Moreover, non-dominated sorting genetic algorithm-II (NSGA-II) was subjected to the GRNN to find the optimal type and level of media and carbohydrate source for obtaining the best seed germination indices (germination rate and mean germination time). According to the optimization process, 0.43 strength mMS medium supplemented with 2.3 % sucrose would result in the best outcomes. This result showed that a moderate level of salts existing in culture media (0.43 strength of mMS medium) supplemented with a moderate level of sucrose (2.3 %) can improve in vitro seed germination of hemp. The results of a validation experiment revealed that there was a negligible difference between the experimental data and the optimized result. Therefore, GRNN-NSGA-II provided an accurate prediction of seed germination and can likely be employed to optimize different factors involved in in vitro culture of this multi-purpose crop.

大麻的体外种子萌发作为植物生命周期的第一个生理阶段，不仅对于研究影响栽培条件的因素很重要，而且对于获得幼体组织作为不同体外程序的潜在外植体也至关重要。另一方面，体外种子萌发是一个多变量的生物过程，可受遗传（基因型）和物理因素（培养基成分和环境条件）的影响。因此，人工神经网络 (ANN) 等强大的数学方法非常适合分析数据并优化这个复杂系统的条件。目前的研究旨在评估不同类型和浓度的碳水化合物来源（蔗糖和葡萄糖）以及不同强度的 DKW（司机和国烧核桃）和 mMS（Murashige 和 Skoog 培养基，范德萨尔姆改良）培养基对种子萌发指数及体外形态特征-通过使用三种人工神经网络，包括多层感知器 (MLP)、径向基函数 (RBF) 和广义回归神经网络 (GRNN) 来种植大麻幼苗。GRNN 模型显示出更高的预测准确度 ( r ²>0.70) 在所有发芽指数和形态特征的训练和测试集中，与 RBF 或 MLP 相比。此外，非支配排序遗传算法-II (NSGA-II) 受到 GRNN 的影响，以寻找培养基和碳水化合物源的最佳类型和水平，以获得最佳种子发芽指数（发芽率和平均发芽时间）。根据优化过程，添加 2.3% 蔗糖的 0.43 强度 mMS 培养基将产生最佳结果。该结果表明，培养基中存在中等水平的盐（0.43 浓度的 mMS 培养基）并补充有中等水平的蔗糖（2.3 %）可以在体外改善大麻种子发芽。验证实验的结果表明，实验数据与优化结果之间的差异可以忽略不计。因此，GRNN-NSGA-II 提供了种子发芽的准确预测，并且可能用于优化这种多用途作物体外培养中涉及的不同因素。