DNNGP, a deep neural network-based method for genomic prediction using multi-omics data in plants.

Area of Science:

• Plant breeding and genetics
• Bioinformatics and computational biology
• Genomics and multi-omics data analysis

Background:

• Traditional genomic prediction models struggle with complex genotype-phenotype relationships.
• Non-linear models, like deep neural networks, show promise for capturing non-additive genetic effects.
• Integrating multi-omics data is crucial for enhancing plant trait improvement.

Purpose of the Study:

• To introduce deep neural network genomic prediction (DNNGP), a novel deep learning method for plant genomic prediction.
• To evaluate DNNGP's performance in integrating multi-omics data for predicting agronomic traits.
• To compare DNNGP against established genomic selection methods.

Main Methods:

• Developed DNNGP, a deep learning model with a multilayered hierarchical structure.
• Trained and tested DNNGP on four diverse plant datasets.
• Compared DNNGP's prediction accuracy, computational time, and robustness against GBLUP, LightGBM, SVR, DeepGS, and DLGWAS.

Main Results:

• DNNGP demonstrated competitive performance on small datasets and superior prediction accuracy on large-scale breeding data.
• The method effectively integrates multi-omics data, learning features dynamically and avoiding overfitting.
• DNNGP showed comparable computation time to common methods and was up to 10 times faster than DeepGS.

Conclusions:

• DNNGP is a superior genomic selection method compared to existing approaches.
• The model offers a practical and robust approach for integrating complex and large multi-omics datasets into plant breeding platforms.
• DNNGP facilitates accelerated agronomic trait improvement through enhanced genomic prediction.