Scholarlink

<p>Somatic embryogenesis (SE) is an important technique for clonal propagation and conservation of <i>Juglans regia</i> L.; however, its efficiency is limited by low embryo maturation and germination rates. Traditional optimization approaches often fail to capture the complex and nonlinear interactions among plant growth regulators (PGRs) and culture conditions. This study aimed to integrate machine learning (ML) modeling with multi-objective optimization to enhance SE development in Persian walnut cv. Chandler. Four ML algorithms (k-nearest neighbors (KNN), support vector machine (SVM), AdaBoost, and CatBoost) were evaluated for their ability to predict embryo size, maturation percentage, and germination percentage based on eight culture parameters. Among them, CatBoost demonstrated the highest predictive accuracy (R² = 0.95, 0.97, and 0.67 for embryo size, maturation, and germination, respectively) and the lowest error metrics across all traits. Using CatBoost as the predictive engine in combination with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II), an optimal culture protocol was identified for maximizing germination. The protocol consisted of basal DKW medium supplemented with 21.73&#xa0;g L⁻¹ sucrose, 2.72&#xa0;mg L⁻¹ abscisic acid (ABA), 4.06% polyethylene glycol (PEG), 2.48&#xa0;mg L⁻¹ gibberellic acid (GA₃), 0.11&#xa0;mg L⁻¹ 6-benzylaminopurine (BAP), and 0.70&#xa0;mg L⁻¹ kinetin (Kin), combined with 50 days of chilling. This optimized protocol was predicted to yield 37.5% germination. Additionally, multi-objective optimization predicted a DKW-based medium containing 1.90&#xa0;mg L⁻¹ ABA, 43.67&#xa0;g L⁻¹ sucrose, 7.38% PEG, and 21 days of chilling, resulting in 85.85% maturation and an embryo size score of 3.93 on a visual five-point scale. Experimental validation demonstrated that the optimized DKW-based medium outperformed the standard DKW, MS, and WPM media, highlighting the effectiveness of the optimization process. Overall, the integration of CatBoost modeling with NSGA-II optimization provides a robust, data-driven framework for improving walnut SE. This approach elucidates biologically meaningful interactions among culture variables and offers a transferable strategy for enhancing in vitro propagation of other recalcitrant woody species.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning and multi-objective optimization framework for improving somatic embryogenesis in Persian walnut (Juglans regia L.)

  • 期刊论文

  • 发表日期：

    2026年6月26日

查看全文
  • Zahra Makari Deligani,
  • Mohammad M. Arab,
  • Mehdi Younessi-Hamzekhanlu,
  • Mohammad Soltani,
  • Maliheh Eftekhari,
  • Kourosh Vahdati

摘要

Somatic embryogenesis (SE) is an important technique for clonal propagation and conservation of Juglans regia L.; however, its efficiency is limited by low embryo maturation and germination rates. Traditional optimization approaches often fail to capture the complex and nonlinear interactions among plant growth regulators (PGRs) and culture conditions. This study aimed to integrate machine learning (ML) modeling with multi-objective optimization to enhance SE development in Persian walnut cv. Chandler. Four ML algorithms (k-nearest neighbors (KNN), support vector machine (SVM), AdaBoost, and CatBoost) were evaluated for their ability to predict embryo size, maturation percentage, and germination percentage based on eight culture parameters. Among them, CatBoost demonstrated the highest predictive accuracy (R² = 0.95, 0.97, and 0.67 for embryo size, maturation, and germination, respectively) and the lowest error metrics across all traits. Using CatBoost as the predictive engine in combination with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II), an optimal culture protocol was identified for maximizing germination. The protocol consisted of basal DKW medium supplemented with 21.73 g L⁻¹ sucrose, 2.72 mg L⁻¹ abscisic acid (ABA), 4.06% polyethylene glycol (PEG), 2.48 mg L⁻¹ gibberellic acid (GA₃), 0.11 mg L⁻¹ 6-benzylaminopurine (BAP), and 0.70 mg L⁻¹ kinetin (Kin), combined with 50 days of chilling. This optimized protocol was predicted to yield 37.5% germination. Additionally, multi-objective optimization predicted a DKW-based medium containing 1.90 mg L⁻¹ ABA, 43.67 g L⁻¹ sucrose, 7.38% PEG, and 21 days of chilling, resulting in 85.85% maturation and an embryo size score of 3.93 on a visual five-point scale. Experimental validation demonstrated that the optimized DKW-based medium outperformed the standard DKW, MS, and WPM media, highlighting the effectiveness of the optimization process. Overall, the integration of CatBoost modeling with NSGA-II optimization provides a robust, data-driven framework for improving walnut SE. This approach elucidates biologically meaningful interactions among culture variables and offers a transferable strategy for enhancing in vitro propagation of other recalcitrant woody species.