WheatGP, a genomic prediction method based on CNN and LSTM.

Area of Science:

• Agricultural Science
• Genomics
• Bioinformatics

Background:

• Wheat breeding faces challenges due to complex genetics and trait variation, impacting food security.
• Accurate prediction of desirable traits is crucial for developing superior wheat varieties.
• Existing genomic prediction methods may not fully capture complex genetic interactions.

Purpose of the Study:

• To propose and evaluate a novel genomic prediction method for wheat, named WheatGP.
• To improve phenotype prediction accuracy by incorporating additive and epistatic genetic effects.
• To provide a high-performance tool for efficient and optimized wheat breeding.

Main Methods:

• WheatGP utilizes a hybrid deep learning architecture combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) modules.
• CNNs capture short-range genomic sequence dependencies, while LSTMs model long-distance gene locus relationships.
• SHapley Additive exPlanations (SHAP) were used to interpret model predictions.

Main Results:

• WheatGP significantly outperforms traditional methods like rrBLUP, XGBoost, SVR, and DNNGP in prediction accuracy.
• Achieved prediction accuracy for wheat yield was 0.73, with accuracies for other agronomic traits ranging from 0.62 to 0.78.
• The model demonstrated robust performance across different crop types and multi-omics datasets.

Conclusions:

• WheatGP offers a powerful approach for comprehensive feature extraction from genomic data.
• This method enhances genomic prediction accuracy, paving the way for accelerated and optimized wheat breeding.
• WheatGP holds potential for advancing crop improvement strategies beyond wheat.