Related Experiment Video
Updated: Jun 17, 2025
09:43
Author Spotlight: Streamlining Rice Breeding with CRISPR/Cas for Obtaining Optimal Phenotypic and Agronomic Traits
Published on: January 3, 2025
2.3K
Evaluation of deep learning for predicting rice traits using structural and single-nucleotide genomic variants
Ioanna-Theoni Vourlaki1,2, Sebastián E Ramos-Onsins3, Miguel Pérez-Enciso3,4,5
1Centre for Research in Agricultural Genomics CSIC-IRTA-UAB-UB, Campus UAB, Edifici CRAG, Bellaterra, 08193, Barcelona, Spain. ioanna.vourlaki@irta.cat.
Plant Methods
|August 10, 2024
Summary
Structural genomic variants (SVs) improve trait prediction in rice, especially when combined with Single Nucleotide Polymorphisms (SNPs). Deep Learning (DL) models show superior performance over traditional Bayesian methods for both binary and quantitative traits.
Area of Science:
- Plant genomics
- Quantitative genetics
- Bioinformatics
Background:
- Structural genomic variants (SVs) are key drivers of plant evolution and domestication, contributing significant genomic and phenotypic variability.
- Current crop improvement methods, like genomic prediction, primarily utilize Single Nucleotide Polymorphisms (SNPs), overlooking the potential of SVs.
- Deep Learning (DL) offers a promising approach for genomic prediction, but its efficacy with SVs and SNPs as genetic markers requires investigation.
Purpose of the Study:
- To assess if combining SVs and SNPs enhances trait prediction accuracy in rice compared to using SNPs alone.
- To compare the performance of Deep Learning (DL) models against traditional Bayesian Linear models for genomic prediction.
- To explore the predictive capabilities of different DL architectures (Multilayer Perceptron, Convolutional Neural Network) using various marker input strategies.
Main Methods:
- Comparative analysis of prediction models: BayesC, Bayesian Reproducible Kernel Hilbert Space (RKHS) regression, Multilayer Perceptron, and Convolutional Neural Network.
- Utilized a combination of Structural Genomic Variants (SVs) and Single Nucleotide Polymorphisms (SNPs) as genetic markers.
- Evaluated model performance across four traits and two validation strategies in rice.
Main Results:
- Integrating SVs with SNPs slightly improved prediction ability for complex traits in 87% of cases.
- Deep Learning (DL) models outperformed Bayesian models in 75% of the evaluated scenarios.
- DL models demonstrated a consistent improvement in predicting binary traits compared to Bayesian models.
Conclusions:
- The inclusion of structural genomic variants (SVs) enhances trait prediction in rice, irrespective of the prediction methodology.
- Deep Learning (DL) networks show potential for superior performance over Bayesian models, particularly for binary traits and when training and target sets are dissimilar.
- This study underscores the value of incorporating SVs alongside SNPs in genomic selection for improved crop breeding strategies.
