To improve our previous protocol, we hypothesize that optimizing the disinfection protocol and seed scarification would increase the speed and frequency of seed germination. As with most aspects of a tissue culture system, in vitro disinfection is a complex and non-linear process that is affected by numerous factors such as disinfectant and contaminant types and levels, media pH, immersion time, temperature, and theirinteractions . In the disinfection process, the concentration of disinfectants plays a conflicting dual role relating to contamination frequency and seed viability . Higher disinfectant concentrations generally lead to a greater control over contaminants; however, lower seedling viability is often the trade-off . Therefore, it is necessary to optimize the disinfection process. The disinfection process cannot be represented by a simple stepwise algorithm, especially when the datasets are highly imbalanced and noisy . Therefore, artifificial intelligence models combined with optimization algorithms such as a genetic algorithm can be employed as an efficient and reliable computational method to inter-pret, forecast, and optimize this complex system .

This strategy has been successfully used for modeling and optimizing different tissue culture systems, including in vitro decontamination, shoot proliferation, androgenesis, somatic embryogenesis, secondary metabolite production, and rhizogenesis . Ivashchuk et al. employed multilayer perceptron and radial basis function as two well-known artificial neural networks for modeling and predicting the effect of different disinfectants and immersion times for Bellevalia sarmatica, Echinacea purpurea, and Nigella damascene explant decontamination. They reported that both algorithms were able to accurately model and predict the disinfection process . In another study, Hesami et al. applied a hybrid MLP and non-dominated sorting genetic algorithm-II for the modeling and optimization of disinfectants and immersion times for chrysanthemum leaf segment decontamination. A generalized regression neural network is another type of ANNs that has successfully been used for modeling and predicting different tissue culture processes . Although there exist no reports using GRNN for the modeling and optimization of disinfection process, we previously showed that GRNN has a higher predictive performance than RBF, MLP, and the adaptive neuro-fuzzy inference system for cannabis micropropagation .

Therefore, in the current study, we used GRNN-GA to model and optimize cannabis seed disinfection. Mature seed germination can sometimes be more difficult than immature seed germination due to the increase in the seed coat’s impermeability and the accumulation of inhibitors during seed maturation . Hence, dormancy breaking plays a critical role relating to the speed and frequency of seed germination due to morpho-physiological dormancy . Although there are no reports on the effects of scarification on cannabis seed germination, the positive impact of dormancy breaking by scarification has previously been suggested in several plants, such as Limodorum , Salvia stenophylla , and legumes . Based on this evidence, studying the effect of scarification on cannabis seed germination can pave the way for devising an in vitro seed germination protocol with high speed and germination frequency. The current study uses GRNN-GA to model and optimize cannabis seed disinfection, and investigates the effect of scarification on seed germination. By combining these procedures, a superior in vitro cannabis seed germination protocol that limits contamination while allowing high germination rates in a short timeframe was established. In vitro seed germination of cannabis has great potential to improve the efficiency of elite cultivar selection, pheno-hunting, phenotyping, and to support various in vitro culture methods as initial explant materials .

In orthodox seeds, germination typically initiates with the passive uptake of water by the dry mature seed, and terminates with radicle protrusion through the seed envelope . Different abiotic factors affect seed germination,mainly through regulating the signaling and metabolism pathways of abscisic acid and gibberellic acid . Although cannabis seeds easily germinate within several days under greenhouse or fifield conditions, in vitro cannabis seed germination tends to be more difficult, with lower germination rates spread over a longer period of time. The cause of this difference is unknown, but is likely related to the disinfection protocol that may stress the developing embryo or potentially eliminate microbes that play a role in the germination process. As such, optimizing sterilization and scarification protocols can be considered the two most important procedures for successful in vitro seed germination . The surface sterilization of initial source material, including seeds, is a prerequisite for the success of the culture . Therefore, it is vital to optimize the sterilization protocol while allowing it to remain simple, cheap, environmentally friendly, and efficient .