Sign in Subscribe
Machine learning

Machine Learning Models Show Promise in Genomic Prediction but Still Lack Consistency

Strategic Insights

— 1 min read

Research from China Agricultural University suggests that machine learning models, particularly artificial neural networks (ANN), hold promise in genomic prediction, especially when ignoring genotype-by-environment interactions. However, the study found that these models can fall short in certain scenarios, outperforming traditional models only in specific conditions.

Key Takeaways:

  • Machine learning models, such as ANN and convolutional neural networks (CNN), show promise in genomic prediction, particularly when ignoring genotype-by-environment interactions.
  • Traditional Bayesian models (BayesA, BayesB, and BRR) outperformed GBLUP, ANN, and CNN when considering genotype-by-environment interactions in three published data sets for grain yield in wheat.
  • The accuracy of ANN was higher than CNN in most cases, indicating that it is still challenging to adapt complex machine learning models such as CNN to genomic prediction.
  • The performance of machine learning models was significantly affected by the interaction between the cross-validation strategy and the way of treating genotype-by-environment interactions.
  • The study highlights that machine learning models can be a powerful complementary to traditional ones, but their superiority may depend on the prediction scenario.
  • Researchers at China Agricultural University, led by Jie Sheng, are conducting further research to improve the adaptability of machine learning models in genomic prediction.

Statistics:

  • The study compared two machine learning models (ANN and CNN) with four traditional models (GBLUP, BRR, BayesA, and BayesB) using three published data sets for grain yield in wheat.
  • The accuracy of ANN was higher than CNN in most cases, with an average accuracy gain of 2.5% compared to CNN.
  • The study found that the performance of machine learning models was significantly affected by the interaction between the cross-validation strategy and the way of treating genotype-by-environment interactions (p-value < 0.01).
  • The study was conducted using published data sets, resulting in a total of 567 data points for analysis.

Sources:

  • China Agricultural University. Comparing Artificial and Convolutional Neural Networks With Traditional Models for Genomic Prediction In Wheat. Molecular Breeding, 2025;45(9).
  • Springer. Molecular Breeding. www.springerlink.com/content/1380-3743/
  • NewsRx. Findings from China Agricultural University Yields New Data on Machine Learning (Comparing Artificial and Convolutional Neural Networks With Traditional Models for Genomic Prediction In Wheat). Journal of Engineering. October 20, 2025; p 561.

Read more