Application artificial neural network for predicting chilling resistance of tomato seedlings following drought stress pretreatment

Introduction Exposure of tropical and subtropical plants, such as tomato, to low temperatures may stunt the plant’s growth, induce wilting and necrotic lesions on leaves, and increase susceptibility to diseases and pathogens (Häll‌green and Öquest, 1990). The symptoms of stress-induced injuries in these plants appear from 48 to 72 h, however, this duration varies from species to species and also depends upon the plant sensitivity to chilling stress. Various phenotyp‌ic symptoms in response to chilling stress include reduced leaf expansion, wilting as well as chlorosis (yellowing of leaves) and may lead to necrosis (death of tissue). Chill‌ing also severely hampers the reproductive development of some plants (Mahajan and Tuteja, 2005). A strategy, which has recently been investigated in plants under stress condition, is cross-resistance, i.e., exposure of tissue to stress conditions often induces resistance to other stresses. For example, salt stress stimulates cold hardiness in potato and spinach seedlings (Ryu et al., 1995). Various mechanisms explaining the phenomenon of cross-resistance have been proposed and often, although not always, it has been suggested that specific proteins are induced by one type of stress and these are involved in the protection against other types of stress. Artificial Neural Networks (ANN) are as an analytical alternative to conventional modeling techniques, which are frequently limited by strict assumptions of normality, linearity, homogeneity, and variable independence (Salehi, 2014). Neural network model was used for potato storage process modeling (Abdulquadri Oluwo et al., 2013), detecting chilling injury in red delicious apple (ElMasry et al., 2009), to detect defects (leaf roller, bitter pit, russet, puncture and bruises) in Empire and Golden Delicious apples(Kavdır and Guyer, 2004) and Lettuce (Lactuca sativa L.) yield prediction under water stress (Kizil et al., 2012) . In this study, artificial neural network modeling was used to predicting chilling resistance of tomato seedlings following imposing drought stress pretreatment with application 0, 10 and 20% poly ethylene glycol (PEG). Materials and Methods Tomato seeds, cultivar Falaat CH, which is one of the most important cultivars grown in Hamedan, were disinfected in 1% (active ingredient) sodium hypochlorite solution for 10 min to eliminate possible seed-borne microorganisms, then they were rinsed for 1 min under running water prior to dry‌ing for 30 min at room temperature. After that, Seeds were planted into plastic pots filled with a 2:1 mixture of coco peat: perlite. The pots were then transferred to the greenhouse with av‌erage temperature of 25.5/19.5°C (day/night) and natural light. When the seedlings developed four true leaves, they were pretreated with 10 or 20% PEG for 7 days or not. After drought, the seedlings were subjected to chilling 6 h/day at 3°C for 6 days. All plants were assessed 72 h after the end of chilling stress to determine the extent of chilling injury and data were collected. In present study, data were collected from experiments and then all data were randomly divided into 3 partitions: training (40%), validating (20%), and testing data (40%). The testing data was used for estimating the performance of the trained network on new data. The Neurosolution software 6.01 (USA) was used for designing the ANN model. In order to predicting chilling effects on tomato seedling attributes, multi-layer perception neural network with 2 input (drought stress and chilling stress effects) and 8 outputs (chlorophyll a, chlorophyll b, total phenol, relative water content, root electrolyte lekage, F0, Fm and proline) was used. Results and Discussion The results showed that the ANN with 7 hidden neurons had the minimum mean absolute error values and high correlation coefficients. The overall agreement between ANN predictions and experimental data was also significant (r=0.92). A plot of the MSE and the number of epochs is shown in Fig. 2. A sharp drop was observed for MSE in the first little iteration (fast training) and training was completed after about 9 epochs. This is a well-known characteristic of the LM optimization method (Salehi and Razavi, 2012). Feed-forward back-propagation ANN models were developed by ElMasry et al. (2009) to investigate the ability of hyperspectral imaging and ANN techniques for the detection of chilling injury in Red apples. They reported that classification accuracy of above 90% was obtained with the use of selected five optimal wavelengths. In another study neural network models was used to predict shelf life of greenhouse lettuce by Lin and Block (2009). Using 2-stage neural network models, an R2 of 0.61 could be achieved for predicting remaining shelf life. This study indicated that neural network modeling has potential for predicting chilling tolerance of tomato seedlings following imposing drought stress pretreatment.