Now, Kelin Wang from the Chinese Academy of Agricultural Sciences and colleagues have developed a deep learning genomic prediction method, Deep Neural Network Genomic Prediction (DNNGP), that incorporates multi-omics data to predict agronomic traits. The DNNGP method has multilayered, hierarchical structures that stack multiple linear and non-linear processing units, which enables dynamic learning features from raw data. DNNGP also employs a callback function to direct early stopping to avoid over-fitting, as well as a non-saturating and non-linear activation function to accelerate the learning process. Wang and colleagues compared DNNGP with five classic genomic selection models on four datasets including wheat599, maize1404, wheat2000 and tomato332. The team found that the new deep learning method performed as well as or better than other models on all datasets in a broad range of prediction tasks. The prediction accuracy of DNNGP is greater than the other methods when large-scale breeding data were used. Additionally, DNNGP can handle complex inputs such as SNP, transcriptomic and proteomic data, making it a practical approach to integrate into the current genomic selection platforms.
Genomic selection has been successfully applied to crops including wheat and rice to improve genetic gains. Currently, increasing the statistical power of genomic prediction methods is an area of active research to help optimize the design of genomic selection experiments. Here, the DNNGP method developed by Wang and colleagues showcases the recent development in genomic prediction methods and can accelerate genomics applications in breeding programmes in the future.
This is a preview of subscription content, access via your institution