Prediction of Multiple-Trait and Multiple-Environment Genomic Data Using Recommender Systems.

Area of Science:

• Genomics
• Plant Breeding
• Bioinformatics

Background:

• Genomic-enabled prediction accuracy is challenged by sparse data and low trait correlations.
• Current genomic selection methods face computational inefficiencies with large datasets.
• Complex statistical models are often computationally impractical for breeders.

Purpose of the Study:

• To explore recommender systems, specifically item-based collaborative filtering (IBCF) and matrix factorization (MF), for genomic prediction.
• To evaluate the performance of IBCF and MF against conventional methods using simulated and real data.
• To assess the computational feasibility and prediction accuracy of these methods in multi-trait, multi-environment contexts.

Main Methods:

• Implemented item-based collaborative filtering (IBCF) and matrix factorization (MF) algorithms.
• Compared IBCF and MF with two conventional genomic prediction methods.
• Utilized both simulated and real-world plant breeding datasets for evaluation.

Main Results:

• Item-based collaborative filtering (IBCF) demonstrated slightly superior prediction accuracy compared to conventional methods and matrix factorization (MF).
• IBCF performance was particularly strong when correlations between environment-trait combinations were moderately high.
• Both IBCF and MF proved computationally feasible for large-scale genomic datasets.

Conclusions:

• IBCF is a promising and computationally efficient approach for enhancing genomic-enabled prediction accuracy in plant breeding.
• The IBCF technique offers practical advantages for breeders managing extensive datasets and complex trait/environment interactions.
• Recommender systems provide a viable alternative to traditional methods for improving prediction models in genomics.