Related Experiment Video
Updated: Jun 26, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.4K
Genomic prediction for sugarcane diseases including hybrid Bayesian-machine learning approaches
Chensong Chen1, Shamsul A Bhuiyan2,3, Elizabeth Ross1
1Center for Animal Science, The Queensland Alliance for Agriculture and Food Innovation, The University of Queensland, Brisbane, QLD, Australia.
Frontiers in Plant Science
|May 16, 2024
Summary
Genomic selection methods, including Bayesian and hybrid approaches with attention networks, effectively predict sugarcane disease resistance. These advanced techniques accelerate the identification of resistant sugarcane clones for improved crop yields.
Area of Science:
- Plant breeding
- Genomics
- Agricultural science
Background:
- Sugarcane diseases like smut and Pachymetra root rot cause significant yield losses exceeding 30%.
- Heritability of disease resistance in sugarcane suggests potential for genetic improvement through selection.
- Genomic selection offers a pathway to accelerate breeding by enabling early identification of resistant seedlings.
Purpose of the Study:
- To evaluate various genomic prediction algorithms for disease resistance in sugarcane clones.
- To compare the accuracy and reliability of different prediction methods, including GBLUP, Bayesian, and machine learning approaches.
- To assess the utility of hybrid methods combining marker selection with attention networks for capturing epistasis.
Main Methods:
- Evaluated GBLUP (with dominance/epistasis), Bayesian (BayesC, BayesR), and machine learning (Random Forest, MLP, CNN, attention networks).
- Developed and tested hybrid methods using BayesR/GWAS for marker selection followed by attention networks.
- Applied random and structured PCA-based cross-validation on 4702 sugarcane clones with genomic and phenotypic data (26k SNP markers).
Main Results:
- Bayesian methods (BayesR, BayesC) demonstrated the highest prediction accuracy.
- Hybrid methods incorporating attention networks closely followed in accuracy and showed the lowest variation across validation folds (lowest MSE).
- Attention networks effectively captured epistasis, particularly in hybrid approaches with selected markers.
Conclusions:
- Bayesian methods are highly effective for genomic prediction of sugarcane disease resistance.
- Hybrid methods utilizing attention mechanisms show promise for predicting clonal performance, especially when non-additive genetic effects are significant.
- The reduced variation in accuracy suggests hybrid methods with attention networks could be valuable for practical sugarcane breeding programs.
