Scholarlink

Key message <p>Integration of conventional bioinformatics approaches with advanced machine learning and explainable AI identified 45 candidate genes and 21 top features (10 positive and 11 negative regulations) influencing heat stress tolerance.</p> Abstract <p>Heat stress is a significant threat to <i>Brassica napus</i> cultivation, a globally important oilseed crop for vegetable oil production. However, identifying the genes associated with heat stress tolerance is challenging using large-scale transcriptomic datasets with traditional approaches. This study combined conventional bioinformatics approaches with advanced machine learning (ML) models to elucidate the heat tolerance mechanism in the seed, flower, leaf, and silique. A total of 1,179 differentially expressed genes (DEGs) were identified, primarily related to detoxification, protein folding, response to heat, and heat shock protein binding, as well as pathways such as alpha-linolenic acid metabolism, glutathione metabolism, and phenylpropanoid biosynthesis. In addition, WGCNA identified 45 candidate hub genes across three modules, associated with four tissues. Interestingly, we trained three ML models, of which the random forest (RF) showed higher performance in terms of ROC (0.98) and accuracy (0.89) than the other two models. Further, we utilized explainable ML, applying SHAP analysis of RF model, and ranked 21 top features (genes) influencing heat stress tolerance, including 10 positive and 11 negative regulators. Among 21 top features, 13 overlapped with traditional bioinformatics approaches (DEGs and WGCNA), whereas eight (38.09%) were uniquely detected via ML models. Finally, expression of seven positive and one negative regulator were validated through RT-qPCR, supporting the findings of integrative meta-transcriptomic and ML results. Our findings provide the valuable resource and highlight the power of ML in genomics to predict the key regulators involved in heat resilience, providing valuable insights into the underlying mechanism of heat stress.</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

Decoding core molecular mechanisms of heat-stress tolerance in Brassica napus using transcriptomics and machine learning

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Muhammad Ikram,
  • Muhammad Farhan,
  • Behnam Derakhshani,
  • Sunjeet Kumar,
  • Aliya Ayaz,
  • Naveed Khan,
  • Enerand Mackon,
  • Babar Usman,
  • Si Chengcheng,
  • Pingwu Liu

摘要

Key message

Integration of conventional bioinformatics approaches with advanced machine learning and explainable AI identified 45 candidate genes and 21 top features (10 positive and 11 negative regulations) influencing heat stress tolerance.

Abstract

Heat stress is a significant threat to Brassica napus cultivation, a globally important oilseed crop for vegetable oil production. However, identifying the genes associated with heat stress tolerance is challenging using large-scale transcriptomic datasets with traditional approaches. This study combined conventional bioinformatics approaches with advanced machine learning (ML) models to elucidate the heat tolerance mechanism in the seed, flower, leaf, and silique. A total of 1,179 differentially expressed genes (DEGs) were identified, primarily related to detoxification, protein folding, response to heat, and heat shock protein binding, as well as pathways such as alpha-linolenic acid metabolism, glutathione metabolism, and phenylpropanoid biosynthesis. In addition, WGCNA identified 45 candidate hub genes across three modules, associated with four tissues. Interestingly, we trained three ML models, of which the random forest (RF) showed higher performance in terms of ROC (0.98) and accuracy (0.89) than the other two models. Further, we utilized explainable ML, applying SHAP analysis of RF model, and ranked 21 top features (genes) influencing heat stress tolerance, including 10 positive and 11 negative regulators. Among 21 top features, 13 overlapped with traditional bioinformatics approaches (DEGs and WGCNA), whereas eight (38.09%) were uniquely detected via ML models. Finally, expression of seven positive and one negative regulator were validated through RT-qPCR, supporting the findings of integrative meta-transcriptomic and ML results. Our findings provide the valuable resource and highlight the power of ML in genomics to predict the key regulators involved in heat resilience, providing valuable insights into the underlying mechanism of heat stress.