Boosting Genomic Prediction Transferability with Sparse Testing.

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Genomic prediction (GP) efficiency is critical for large-scale breeding programs.
• Sparse testing strategies are needed to reduce costs and logistical challenges.
• Refining genomic selection (GS) under sparse testing is an active area of research.

Purpose of the Study:

• To evaluate a sparse testing approach for predicting line performance in untested environments.
• To assess the impact of incorporating external training data on prediction accuracy.
• To determine the optimal use of geographically and temporally related data in GP.

Main Methods:

• Utilized training data from CIMMYT (Obregon, Mexico) and partial data from India.
• Employed a sparse testing strategy to predict line performance in India using Mexican observations.
• Analyzed the effect of training set composition on prediction accuracy.

Main Results:

• Incorporating Obregon data into the training set significantly improved prediction accuracy.
• Prediction accuracy gains were greater with temporally closer data.
• Pearson's correlation improved by over 219% with 50% testing proportion; top line identification also saw substantial gains.

Conclusions:

• Enriching training data with relevant, temporally proximate information is key to enhancing GP performance.
• Unrelated or temporally distant data can decrease prediction accuracy.
• Strategic data integration in sparse testing is vital for efficient breeding programs.