نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشجو کارشناسی ارشد، گروه تولیدات گیاهی، دانشکده علوم کشاورزی و منابع طبیعی، دانشگاه گنبدکاووس، ایران.
2 استادیار، گروه تولیدات گیاهی، دانشکده علوم کشاورزی و منابع طبیعی، دانشگاه گنبدکاووس، ایران.
3 استاد، گروه تولیدات گیاهی، دانشکده علوم کشاورزی و منابع طبیعی، دانشگاه گنبدکاووس، ایران.
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
Objective
Iron toxicity is a major constraint to rice production in acidic flooded soils. Conventional methods for evaluating tolerance to this stress rely on costly and time-consuming phenotyping, which limits their efficiency in breeding programs. This study aimed to develop an efficient cross-condition prediction framework to assess rice tolerance to iron toxicity using phenotypic and genotypic data collected under both normal and stress conditions.
Materials and Methods
Phenotypic and genotypic data derived from iPBS, IRAP, ISSR, and SSR markers were collected from 96 rice inbred lines and their parents under both normal and iron-toxicity stress conditions. Following data preprocessing, including removal of missing values, outlier detection, and normalization, feature selection was performed using three algorithms: F-test, MRMR, and RReliefF. Eleven machine-learning models, including artificial neural network (ANN), support vector machine (SVM), decision tree (DT), random forest (RF), XGBoost, AdaBoost, Gaussian process (GP), linear regression (LR), linear discriminant analysis (LDA), k-nearest neighbors (KNN), and Naive Bayes (NB), were evaluated across twelve prediction scenarios defined by data type and training–testing conditions. Model performance was assessed using MAE, RMSE, and R2R^2R2. The dataset was split into training and test sets at a 70:30 ratio, cross-validation was applied to improve model robustness, and hyperparameters were optimized using grid search and Bayesian optimization.
Results
Prediction scenarios based on full data integration (D10 and D12) showed the highest predictive accuracy, with coefficients of determination exceeding 0.96. Scenario D8, which relied solely on genotypic data from normal conditions and MRMR-based feature selection, achieved moderate predictive accuracy (R2=0.678R^2 = 0.678R2=0.678). This result suggests that marker-derived patterns associated with genotypes’ responses to iron toxicity can be partially captured from normal-condition data and used for preliminary cross-condition screening.
Conclusion
The results indicate that a stepwise prediction strategy can improve the efficiency of preliminary genotype screening for iron-toxicity tolerance. Models such as D8, despite their moderate accuracy, may help reduce costs and optimize resource allocation during early screening stages. However, final selection of tolerant genotypes still requires validation using more accurate models, direct phenotyping under stress conditions, and independent field trials.
کلیدواژهها [English]
)