SoyDNGP: a web-accessible deep learning framework for genomic prediction in soybean breeding

Area of Science:

• Genomics
• Plant Breeding
• Bioinformatics

Background:

• Soybean is a crucial global crop, but its genetic complexity hinders yield and quality optimization.
• Accurate trait prediction tools are essential for advancing soybean breeding programs.
• Existing genomic prediction methods face challenges with complex traits and large datasets.

Purpose of the Study:

• To develop and evaluate SoyDNGP, a novel deep learning model for enhanced soybean genomic prediction.
• To compare SoyDNGP's performance against established methods like DeepGS and DNNGP.
• To assess the model's versatility and applicability across multiple crop species.

Main Methods:

• Development of SoyDNGP, a deep learning architecture for genomic prediction.
• Rigorous performance evaluation focusing on predictive accuracy and model complexity.
• Cross-species validation including cotton, maize, rice, and tomato.
• Creation of a user-friendly web server for accessibility.

Main Results:

• SoyDNGP demonstrated superior predictive accuracy with minimal parameter increase compared to DeepGS and DNNGP.
• The model accurately predicted complex soybean traits across varying sample sizes.
• Consistent and comparable performance was observed in cotton, maize, rice, and tomato, highlighting its versatility.
• A web server was launched, offering trait lookup and prediction functionalities.

Conclusions:

• SoyDNGP offers a significant advancement in genomic prediction for soybean breeding.
• The model's high accuracy and cross-species applicability make it a valuable tool for crop improvement.
• The accessible web server democratizes the use of advanced genomic prediction techniques for researchers and breeders.