Genotype-specific responses to in vitro drought stress in myrtle (Myrtus communis L.): integrating machine learning techniques

Introduction

Myrtle, also known as Myrtus communis L., is a perennial evergreen shrub or small tree native to the Mediterranean region (Cioć et al., 2018). It belongs to the Myrtaceae family, which consists of 100 genera and 3,000 species that grow naturally in tropical and subtropical regions (Rezaee & Kamali, 2014; Şimşek et al., 2023). Various parts of the myrtle plant, including its leaves, flowers, and fruits, contain numerous important constituents for the medical, food, liquor, and cosmetic industries (Serce et al., 2008). These constituents make myrtle useful for anti-genotoxic, anti-mutagenic, antiseptic, anti-inflammatory, and treating internal and tropical infections (Bonjar, 2004). Essential oils derived from myrtle can be utilized in pharmaceutical production, and their leaves are used in fragrance and food industries (Mulas & Cani, 1999). Myrtles can also be consumed as a tea (Flamini et al., 2004; Şimşek et al., 2022). This shrub or small tree typically ranges in height from 0.5 to 3 m, and two subspecies have been identified: M. communis subsp. communis and M. communis subsp. tarentina (Picci & Atzei, 1996).

Myrtles can thrive in various habitats and are consistently the dominant plant in the maquis. They are slow-growing, long-lived evergreen plants that do not shed their leaves during winter (Sumbul et al., 2011). The leaves of the myrtle plant have an aromatic odor, and the fruits are berry-like, typically blackish-purple or white, and ripen during the autumn season (October–December). The ripe fruits are both sugary and astringent and are pollinated by insects. The fruits are covered with a light, waxy layer that helps bees recognize the plant, giving the fruit a matte appearance. The fruits are multi-seeded and have a pleasant, spicy, and aromatic odor. They contain essential oils, tannins, sugars, and organic acids (citric and malic acids) (Aleksic & Knezevic, 2014). The leaves of the myrtle plant also contain small amounts of phenolic acids such as caffeic, ellagic, gallic acid, and quercetin derivatives, as well as large amounts of catechin and myricetin derivatives (Simsek et al., 2020).

Drought is a significant contributor to the decline in plant production. Environmental stress hinders the growth and development of plants by limiting their ability to respond and adapt to their surroundings. The high biological complexity involved in a plant’s response to stress is often accompanied by numerous changes at various levels, including morphological, molecular, cellular, physiological, and biochemical. M. communis has demonstrated remarkable resistance to abiotic stressors; however, the prolonged drought conditions and intense solar radiation experienced during Mediterranean summers may have contributed to its decline (Gucci et al., 1997; Mendes, Gazarini & Rodrigues, 2001).

Plants cope with drought stress by regulating their metabolism through osmosis. This process involves the active accumulation of organic and inorganic compounds, known as osmolytes or osmotic protectors, which occur with decreased cell water potential. The process of osmotic regulation in cells allows osmotic potential to be reduced with the help of osmotic protectors that accumulate at high rates (Sanchez et al., 2004).

Several studies have found that testing plant responses to abiotic stress factors under in vitro conditions is a dependable method that produces consistent results compared to field experiments. This approach allows for the evaluation of many genotypes under stress conditions in a practical manner (Basu, Gangopadhyay & Mukherjee, 2002). For instance, an in vitro study (Di Cori et al., 2013) examined the adaptation mechanisms of myrtle to salinity stress. However, the physiological processes through which M. communis responds to water stress or drought have not been thoroughly documented (Gratani, Catoni & Varone, 2013; Navarro et al., 2009; Tafreshi et al., 2021).

In recent years, in vitro studies have used polyethylene glycol (PEG), sorbitol, and mannitol as abiotic stress factors to create stress conditions, drought agents, pathogens, or phytotoxins directly in the pathogen culture medium as abiotic stress agent (Manoj, Chaudhary & Singh, 2021). In tissue culture, PEG, sorbitol, and mannitol were employed as osmotic stress agents. While PEG is commonly used to create artificial drought conditions by reducing the available water potential in the culture media, it is preferred because it does not cause phytotoxic effects in the culture media (Hassan et al., 2004).

Machine learning (ML) applies data science techniques to address complex challenges across various scientific domains (Bozkurt et al., 2024; Tarraf et al., 2024). Artificial intelligence-based ML models can analyze and predict complex, multivariate datasets (Kul et al., 2020; Isak et al., 2024; Tütüncü et al., 2024) and learn intricate relationships (Jamshidi et al., 2020; Sadat-Hosseini et al., 2022) to overcome these issues. These techniques have gained popularity in recent years for analyzing datasets in multiple fields within plant science. However, machine learning methodologies in plant and agricultural sciences are relatively limited compared to their extensive application in other scientific disciplines (Aasim et al., 2023; Şimşek et al., 2024a). Nevertheless, some researchers have achieved notable success in various areas of plant science, such as plant breeding (van Dijk et al., 2021) and gene function (Mahood, Kruse & Moghe, 2020). Adopting ML techniques offers the advantage of enabling computers to learn independently and convert data into valuable knowledge without human programming (Hesami et al., 2022). Table 1 presents the history of the related literature review.

This study explores the drought stress responses of different myrtle genotypes, hypothesizing that these genotypes display varying levels of tolerance to drought conditions. Myrtle, a highly valued Mediterranean plant with diverse applications, is increasingly confronted with drought due to global climate change. In this regard, the study employs ML analysis utilizing the Gaussian process (GP), support vector machine (SVM), Random Forest (RF), and Extreme Gradient Boosting algorithms (XGBoost) to model and predict micropropagation and rooting efficiency across varying concentrations of polyethylene glycol.

To test this hypothesis, the study aims to: (1) utilize morphological assessments to identify changes related to drought stress; (2) employ ML models to predict drought tolerance based on the observed and morphological data.

Materials and Methods

Results

Machine learning analysis

Table 7 provides a comprehensive evaluation of various ML models in different plant growth prediction tasks and allows for identifying the models with the highest performance metrics. Among these models, determining the models with the highest R² scores and lowest MAE values is crucial in determining the best performer regarding predictive accuracy and precision. Regarding the Micropropagation Rate, the MLP model emerged as the top performer with the highest R² score of 0.58, indicative of its superior predictive capability. Additionally, boasting the lowest MAE of 0.11 further solidifies the MLP’s position as the optimal choice for predicting the micropropagation rate. The ranking of the models in predicting the micropropagation rate is as follows: MLP < GP < SVM < RF < XGBoost. For the prediction of Plant Height, both MLP and XGBoost exhibited equally impressive R² scores of 0.80, suggesting a comparable predictive probability. However, considering the MAE values, MLP outperformed XGBoost with a lower value of 0.06, cementing its superiority in accurately predicting plant height. MLP and XGBoost emerged as the top performers, followed by GP, SVM, and RF, which exhibited comparable performances (MLP < XGBoost < GP = SVM = RF). Similarly, in the task of predicting Plant Height in Rooting Medium, MLP and XGBoost once again shared the highest R² score of 0.79. However, with a lower MAE of 0.08 compared to XGBoost’s 0.09, MLP takes the lead as the preferred model for this specific prediction task. MLP and XGBoost maintained their superiority, followed by GP, SVM, and RF, which demonstrated similar performance levels (MLP < XGBoost < GP = SVM = RF).

Table 7:

Assessment metrics for the machine learning models.

	ML models	R2	MAE	RMSE
Micropropagation rate	MLP	0.58	0.11	0.14
	GP	0.56	0.10	0.13
	SVM	0.55	0.11	0.14
	RF	0.54	0.11	0.14
	XGBoost	0.52	0.12	0.15
Plant height	MLP	0.80	0.06	0.08
	GP	0.79	0.07	0.09
	SVM	0.78	0.06	0.08
	RF	0.78	0.07	0.09
	XGBoost	0.80	0.06	0.09
Plant height in rooting medium	MLP	0.79	0.08	0.10
	GP	0.78	0.08	0.10
	SVM	0.77	0.08	0.10
	RF	0.77	0.08	0.10
	XGBoost	0.78	0.08	0.10
Number of roots	MLP	0.32	0.14	0.17
	GP	0.29	0.15	0.19
	SVM	0.28	0.14	0.17
	RF	0.27	0.15	0.18
	XGBoost	0.24	0.17	0.20
Root length	MLP	0.46	0.13	0.16
	GP	0.41	0.14	0.17
	SVM	0.40	0.15	0.18
	RF	0.39	0.15	0.19
	XGBoost	0.50	0.12	0.15

DOI: 10.7717/peerj.18081/table-7

Note:

ML, Machine Learning; MLP, Multilayer Perceptron; GP, Gaussian Process; SVM, Support Vector Machines; RF, Random Forest; R², Coefficient of Determination; MAE, Mean Absolute Error; RMSE, Root Mean Square Error.

When it comes to predicting the Number of Roots, the multilayer perceptron (MLP) model attained the highest R² score of 0.32. However, its performance in terms of mean absolute error (MAE) was inferior, with a value of 0.14. Conversely, the SVM model demonstrated superior accuracy in predicting the number of roots, with a slightly lower R² score of 0.28 but a lower MAE of 0.14, positioning it as the preferred model. The SVM model emerged as the top choice, followed by MLP, RF, GB, and XGBoost (SVM < MLP < RF < GB < XGBoost). For predicting Root Length, the XGBoost model exhibited the highest R² score of 0.50 and the lowest MAE of 0.12, showcasing its exceptional performance in accurately forecasting root length. XGBoost was the most effective model, followed by MLP, with GB, SVM, and RF exhibiting comparable performance levels (XGBoost < MLP < GB = SVM = RF).

In conclusion, while the MLP model consistently achieved high R² scores across various tasks, its performance in terms of MAE varied. Therefore, it is crucial to consider both the R² scores and MAE values when determining the optimal model for each specific plant growth metric. The actual and predicted values are illustrated in Figs. 1–5, which compares the samples on the horizontal axis with the model’s predicted outcomes on the vertical axis.

Discussion

In discussing our findings on the impact of PEG-induced stress in myrtle, it is pertinent to contextualize our results within the spectrum of existing research. This approach highlights the unique contributions of our study and integrates it into the broader scientific discourse on myrtle’s response to various stressors. Our study’s observation of decreased growth rates and developmental parameters in myrtle under increasing concentrations of PEG aligns with the findings of Tafreshi et al. (2021). They reported a reduction in shoot regeneration and physiological parameters under similar stress conditions, thereby underlining myrtle’s sensitivity to water stress. The novelty of our research lies in its detailed quantification of the specific impact on different genotypes of myrtle, which enriches the understanding of genotype-specific responses to stress. The relation of our findings to the work of Di Cori et al. (2013) and Aghaie, Hosseini Tafreshi & Toghyani (2022) on salt stress is also notable. Both drought stress (induced by PEG in our study) and salt stress affect the osmotic potential around plant roots. The decrease in growth and development we observed under PEG-induced stress parallels the growth reduction seen under salt stress, suggesting a potentially shared osmotic stress response mechanism in myrtle. Intriguingly, our results contrast with those of Azizi et al. (2021), who demonstrated that microbial inoculation could mitigate the effects of water deficit in myrtle. While we observed a clear decline in various growth parameters with increasing PEG concentrations, Azizi et al.’s (2021) findings indicate that certain treatments could ameliorate these negative impacts. This disparity underscores the complexity of myrtle’s response to environmental stress and suggests the potential for developing strategies to enhance myrtle’s resilience. Furthermore, the integration of our findings with propagation studies such as those by Şimşek et al. (2023), Aka Kaçar et al. (2020, 2015), and Şan, Karakurt & Dönmez (2015) is essential. These studies, focusing on myrtle propagation, underline the importance of optimizing growth conditions. The challenges posed by environmental stressors like PEG in tissue culture and propagation efforts highlighted in our study add a critical dimension to this narrative, specifically illustrating how PEG can adversely affect these processes.

In the response of myrtle to PEG-induced drought stress, it is essential to consider the broader context of how drought stress impacts various plant species, particularly in fruit development. This comprehensive understanding aids in drawing parallels and distinctions between our results and those observed in other plants, thereby enriching the insights gained from our study. The research conducted by Zekai et al. (2022) on banana cultivars under in vitro drought stress provides a vital molecular perspective. They observed differential expression levels in the CDPK gene family, integral to drought stress response, among different banana cultivars. This cultivar-specific molecular response is particularly instructive for our study on myrtle. Just as banana cultivars exhibited varied gene expression responses to drought, the different myrtle genotypes in our study may also have distinct molecular and physiological responses to PEG-induced stress. Understanding these responses at a genetic level can be crucial for developing drought-resilient myrtle cultivars. Furthermore, studies on various fruits, such as those by Castellarin et al. (2007) on grapevine, Jiang et al. (2020) on jujube, and Rathnayaka et al. (2021) on chili pepper, have shown that drought stress significantly alters biochemical compositions. These alterations include changes in flavonoids, anthocyanins, sugars, and capsaicinoids, all of which are crucial for fruit quality and nutritional value. Our study on myrtle aligns with these findings, suggesting that PEG-induced drought stress similarly impacts metabolic pathways in myrtle, affecting its growth and development. The work of Asakura et al. (2021) on tomatoes further highlights the diverse impacts of drought stress, demonstrating an increase in fatty acid synthesis and alterations in fruit yield and ripening periods. These changes are reflective of the complex ways in which plants respond biochemically and physiologically to water scarcity. Incorporating these insights into our study, it becomes evident that drought stress, whether natural or simulated, as in the case of PEG, can have profound effects on plant physiology and biochemistry. This comparison with other species underlines the importance of examining such stress impacts in both agricultural and ornamental plants like myrtle. It also suggests that myrtle might exhibit changes in key metabolic pathways under drought conditions similar to other species, which could have implications for its use and cultivation. Devireddy et al. (2021) emphasize the importance of understanding the integration of reactive oxygen species and hormone signaling during abiotic stress. This aspect is also relevant to our study, as these underlying biochemical and molecular mechanisms may also be at play in myrtle’s response to drought stress, influencing its growth and development. Our study on myrtle underscores the complexity of plant responses to drought stress when viewed in conjunction with these broader findings. This comparative analysis deepens our understanding of how different species, including myrtle, react to such stressors and opens up new avenues for research and strategies to mitigate drought stress impacts. Developing a comprehensive approach that considers both genetic and biochemical responses could lead to cultivating plant varieties that are more resilient to drought, enhancing both agricultural productivity and the preservation of ornamental species. In conclusion, our research contributes significantly to the current understanding of myrtle’s response to drought stress, particularly that induced by PEG. By echoing the findings of reduced growth under stress seen in other studies and providing new insights into genotype-specific responses, our study not only situates itself within the larger body of myrtle research but also opens pathways for future investigations, particularly in the realm of developing strategies to bolster myrtle’s resilience against environmental stressors.

In response to drought stress, plants initiate a range of physiological and molecular adaptations to maintain cellular integrity and homeostasis. Our study’s findings on the variation in drought tolerance among myrtle genotypes correlate with the observed differential regulation of key osmoprotective compounds and antioxidant enzymes. This observation aligns with these molecules’ roles in mitigating the detrimental effects of osmotic stress and oxidative damage (Şimşek et al., 2024b). Furthermore, the increased expression of stress-related genes, as seen in our experimental outcomes, mirrors the molecular adaptations described by Tafreshi et al. (2021), who highlighted the upregulation of drought-responsive genes under controlled stress conditions induced by PEG treatments. These findings underscore the importance of a multi-faceted approach to understanding plant drought tolerance, integrating physiological and genetic insights. We can better understand how plants manage water deficit by elucidating the specific physiological pathways involved in drought response, as observed in our study and supported by external literature. This comprehensive perspective is vital for developing strategies to enhance crop resilience in the face of increasing drought incidence due to climate change.

The research presented in our study and the cited literature collectively highlight the growing integration of ML algorithms in plant tissue culture and their significant impact on enhancing the efficiency and predictability of various in vitro processes. This comparative discussion aims to explore the utilization of ML in plant tissue culture, focusing on different studies and their unique contributions to the field.

Our study utilized a comprehensive approach involving tissue culture, rooting trials, and acclimatization processes, emphasizing the genotype’s responses to drought stress. We observed that different genotypes exhibited varying micropropagation rates under increasing concentrations of PEG, with Black Fruited genotypes showing higher rates than White Fruited ones. This aspect of genotype-specific responses is echoed in Rezaei et al. (2023), where the unpredictable and genotype-dependent callogenesis in petunia was addressed using ML. Both studies underscore the importance of genotype selection and optimization in tissue culture, which ML algorithms can efficiently handle. The use of different ML algorithms (MLP, GP, XGBoost, SVM, RF) in our study to model various plant growth parameters showcases the versatility of ML in handling diverse aspects of plant tissue culture. This versatility is similarly demonstrated in García-Pérez et al. (2020), where artificial neural networks were applied to optimize the extraction of phenolic compounds in Bryophyllum species. The diverse ML applications in both studies signify its potential in a wide array of tissue culture processes. Kirtis, Aasim & Katırcı (2022) and Aasim et al. (2022b) both emphasize the role of ML in optimizing in vitro regeneration protocols, a theme that aligns well with our study. The use of ML to predict outcomes in tissue culture, as seen in the successful regeneration of desi chickpea and common bean, parallels our findings, where ML models efficiently predicted the micropropagation rates and other growth parameters. Aasim et al. (2022b) and Jafari & Daneshvar (2023) further expand on the utility of ML in plant tissue culture, focusing on hemp and Passiflora caerulea, respectively. Both studies highlight the role of ML in optimizing germination and regeneration protocols, resonating with our study’s emphasis on the predictive power of ML in tissue culture. Şimşek et al. (2024a) and Jafari & Shahsavar (2020) demonstrate the application of ML in enhancing the efficiency of micropropagation and rooting protocols in lavender and modeling morphological responses of citrus to drought stress, respectively. These studies, along with our own, illustrate the broad applicability of ML in various aspects of plant tissue culture, from micropropagation to stress response modeling. In conclusion, these studies’ collective insights reveal that ML is a powerful tool in plant tissue culture, offering enhanced predictability, optimization, and efficiency in various processes. The ability of ML algorithms to handle complex datasets and provide precise predictions is invaluable, particularly in genotype-specific responses and optimization of growth conditions. The integration of ML in plant tissue culture research represents a significant advancement in the field, paving the way for more controlled, efficient, and productive cultivation practices.

Our study provides a foundation for further research into the response of myrtle genotypes to drought stress, but several areas remain ripe for exploration. Future studies could expand the scope by investigating the effects of other abiotic stressors, such as salinity or temperature extremes, on myrtle’s physiological and genetic responses. Integrating advanced ML models, such as deep learning or ensemble techniques, could enhance our analyses’ accuracy and predictive power. Exploring the molecular mechanisms underlying the observed genotype-specific responses, perhaps through genomic or transcriptomic studies, could offer deeper insights into the resilience traits of myrtle. Furthermore, comparative studies across different plant species could help identify universal stress response mechanisms and contribute to broader agricultural and ecological applications. Such interdisciplinary approaches will broaden our understanding of myrtle’s adaptive capabilities and aid in developing robust strategies for cultivating drought-resistant plant varieties in the face of climate change.

Conclusion

The investigation into the performance of superior myrtle genotypes under in vitro drought conditions has provided valuable insights into the adaptive capabilities of these plants. Our study has shown that myrtle genotypes respond differently to increasing concentrations of polyethylene glycol, a surrogate for drought stress. Particularly, Black Fruited genotypes exhibited higher resilience in terms of micropropagation rates compared to their White Fruited counterparts. The application of machine learning models has been instrumental in predicting and analyzing the performance of these genotypes under stress. The models demonstrated a high degree of accuracy, indicating their potential as reliable tools in the field of plant tissue culture and stress response analysis. This approach enhances our understanding of genotype-specific responses to environmental stresses and paves the way for more efficient and targeted cultivation strategies. Overall, our findings underscore the importance of considering genotype-specific traits in developing cultivation practices for myrtle, especially in the face of increasing drought conditions due to climate change. The successful integration of machine learning into this field offers a promising direction for future research and application, potentially revolutionizing plant tissue culture and breeding strategies for enhanced resilience against environmental stresses.

Supplemental Information