Evaluating Methods of Updating Training Data in Long-Term Genomewide Selection

Area of Science:

• Plant breeding
• Quantitative genetics
• Genomics

Background:

• Genomewide selection (GWS) accelerates genetic gains but prediction accuracy can decrease over breeding cycles due to changing linkage disequilibrium (LD).
• Previous studies highlight the need to update training populations to maintain accuracy, but optimal update strategies remain unexplored.

Purpose of the Study:

• To investigate optimal methods for updating training populations in GWS to sustain prediction accuracy and response to selection.
• To evaluate the impact of different update strategies and data inclusion methods in a barley breeding simulation.

Main Methods:

• A barley (Hordeum vulgare L.) breeding simulation was used to compare updating strategies: best predicted lines, worst predicted lines, both, random lines, criterion-selected lines, or no update.
• Prediction accuracy and response to selection were assessed over multiple breeding cycles.
• The effect of using all historical data versus only recent data in the training population was also examined.

Main Results:

• In the short term, updating with the best predicted lines or both best and worst predicted lines yielded high prediction accuracy and genetic gain.
• Over the long term, all update methods (except no update) showed similar performance.
• Using a smaller, more recent training population provided a slight advantage in prediction accuracy and genetic gain compared to using all data.

Conclusions:

• Updating the training population is crucial for maintaining prediction accuracy in GWS.
• While several update methods are effective in the short term, their long-term performance converges.
• Employing a recent, focused training dataset offers a practical and effective approach for breeders to optimize genetic gains.