Poster # 507
Maggie Gillum 1, Jeanette Lyerly 2, Lauren Brzozowski 1, and David Van Sanford 1
1. University of Kentucky, Department of Plant and Soil Science, Lexington Kentucky
2. USDA-ARS, Department of Crop and Soil Sciences, University of North Carolina, Raleigh, NC, 27695 USA
Corresponding Author: Maggie Gillum, megi249@uky.edu
Gillum, Maggie
Genomic selection offers substantial potential for
early-generation selection in breeding programs, but training population
composition may influence predictions. The goal of this study is to assess the
differences and accuracy of genomic predictions within an elite-by-elite
family, typical of the breeding program, based on different training
populations. To evaluate the impact of training populations on genomic
prediction-based selections, we used two different training populations, KY and
MDX. The KY training population consists of University of Kentucky breeding
lines. The MDX training population uses lines trialed in the regional
Mason-Dixon nursery, consisting of lines from the University of Kentucky,
Virginia Tech, and University of Maryland breeding programs plus several
USDA-ARS breeding lines from the program at Raleigh, North Carolina. A fourth
training population, FHB, was used for deoxynivalenol (DON) genomic
predictions. The FHB training population consists of historical data from the
Southern Uniform Scab Nursery. Genomic estimated breeding values (GEBVs)
for all training populations were derived using rrBLUP. Phenotypic evaluations
were performed on 120 F2:4 soft red winter wheat (Triticum
aestivum L.) lines segregating for Fhb1, 1A Neuse, 4A Neuse, and 1B
Jamestown. Phenotypes for grain yield, test weight, heading date, and plant
height were measured across two Kentucky locations, and DON accumulation was
measured in the inoculated Fusarium head blight nursery. The MDX training population had the
highest prediction accuracy for grain yield, test weight, heading date,
and plant height. The FHB training population had a prediction accuracy of r
= 0.195 for DON accumulation. Using a selection index, the MDX training
population had the highest prediction accuracy (r = 0.242). However, at a
moderate selection intensity (40%), the KY training population was able to
capture more of the top-performing lines than the MDX training population
(46.8% versus 44.7%). Furthermore, binomial tests revealed that the MDX
training population enriched selection for Fhb1 when using the selection index
or predicted yield (P = 0.04). These findings highlight the trade-off of
regional versus program-specific training populations. The regional MDX
training population had higher prediction accuracy, but the program-specific KY
training population was able to capture more of the top-performing lines. These
results indicate the importance of training population composition and that
breeders should optimize training populations for selection objectives to
maximize overall prediction accuracy.
© Copyright
2026 by individual authors. All rights reserved. No part of this abstract or paper publication may be reproduced without prior permission from the applicable author(s).