Document Type : Original Article
Authors
1
Professor, Department of Plant Production, Faculty of Agriculture and Natural Resources, Gonbad Kavous University, Gonbad Kavous, Iran.
2
Assistant Professor, Department of Plant Production, Faculty of Agriculture and Natural Resources, Gonbad Kavous University, Gonbad Kavous, Iran.
3
Department of Agricultural and Horticultural Sciences Research, Mazandaran Agricultural Resources Research and Education Center, Agricultural Research, Education and Extension Organization (AREEO), Sari, Iran.
4
Former PhD Student, Department of Plant Breeding, Faculty of Agriculture Science, Gorgan Agriculture Science and Natural Resource University, Iran.
5
MSc Student, Department of Plant Production, Faculty of Agriculture and Natural Resources, Gonbad Kavous University, Gonbad Kavous, Iran.
Abstract
Objective
This study aimed to predict the response of 297 wheat genotypes to common regional diseases using 71 iPBS and 81 SSR molecular markers.
Materials and Methods
The wheat genotypes were cultivated using an augmented design within a completely randomized layout at the Qarahkil Agricultural Research Station in Qaemshahr. Both molecular and phenotypic data were analyzed. Feature selection and classification algorithms were employed for data analysis, including Random Forest and the InfoGainAttributeEval evaluation model.
Results
The analysis revealed substantial genetic diversity among wheat genotypes at both the molecular and phenotypic levels. However, the effectiveness of the predictive algorithms varied depending on the disease. The algorithms showed limited predictive accuracy for diseases such as tan spot and Fusarium head blight, suggesting that the available data may be insufficient to determine genotype resistance or susceptibility reliably. In contrast, yellow rust demonstrated the highest prediction accuracy, averaging 70%, indicating that the models were highly effective in predicting genotype responses. The algorithms also performed relatively well for powdery mildew and brown rust, achieving average prediction accuracies of 56.00% and 56.28%, respectively. However, prediction accuracy for brown spots was low, averaging 34.90%, indicating limited model performance for this disease.
Further analysis showed that phenotypic data were more effective in predicting responses to powdery mildew, whereas molecular data provided better predictions for brown rust. Overall, the SSR and iPBS markers, combined with the Random Forest classification algorithm and the InfoGainAttributeEval model, enabled accurate prediction of genotype responses for several diseases.
Conclusion
The findings indicate that the predictive power of molecular and phenotypic data varies by disease. While molecular data were more accurate for some diseases, phenotypic data were more effective for others. Notably, both data types accurately predicted genotype responses to yellow rust.
Keywords
)