Abstract Template doc - 54.50 kB

advertisement
Clermont-Ferrand, France, 2013
International Workshop on Wheat Genomic Selection
USEFULNESS OF GENOME-WIDE PREDICTION IN WHEAT BREEDING
PROGRAMMES
Gilles Charmet1, Eric Storlie2
1
INRA-UBP, UMR GDEC, 5 chemin de Beaulieu 63100 Clermont-Ferrand, FRANCE
Colorado States University, Soil and Crop Science, Fort Collins, Colorado 80523, USA
2
Submitted for: oral presentation / poster (choose one)
Selected session : oral session number or poster session
Contact email address: gilles.charmet@clermont.inra.fr
Dear Author, when submitting your abstract you are kindly requested to write it
directly into this sample document and also by filling the form on the website. Upload
the Word document as well when submitting your abstract online. Please kindly note
that abstracts submitted in any other way cannot be accepted. The article must be not
more than 1 page long. pt. 10
With the expected development of thousands of molecular markers in most crops, the
marker-assisted selection theory has recently shifted from the use of a few markers targeted
in QTL regions (or derived from candidate genes) to the use of many more markers covering
the whole genome. Provided that a sufficient level of linkage disequilibrium exists between
the markers used for genotyping and the true genes underlying QTLs for complex traits,
these genome wide markers can be used to predict the true breeding value [1]. To be useful
for breeding purposes, the accuracy of this Genome Estimate of Breeding Value (GEBV)
should not be worse than the estimation based on phenotype, which is not always the best
predictor of Breeding Value, particularly in the presence of GxE interactions. Moreover, since
GEBV allows shorter selection cycles, an overall improvement of genetic progress per time
unit is expected. [2]. We present a case study using DArT markers on the INRA wheat
breeding programme, in an attempt to implement whole genome selection as an alternative to
phenotypic selection. This investigation assesses different models – Pedigree BLUP, Ridge
Regression BLUP, Bayesian Ridge Regression and Bayesian Lasso – ability to predict either
simulated or real data (yield in a multisite, multi-annual network). The prediction coefficients
obtained when the target population is a random sample (i.e. cross-validation) of the data are
of the same magnitude of those reported in the literature. However, when the target
population is a subset of genotypes studied in a given year (out of 10), the prediction quality
is much lower. However, from a practical point of view, the ranking of (best) genotypes is
more important than the prediction accuracy. Again the goodness of fit of best genotypes
ranking is highly variable from year to year. Implication of results for practical implementation
of genomic prediction in breeding programmes is discussed.
1. Meuwissen THE, Hayes B, Goddard ME (2001) Prediction of total genetic value using
genome-wide dense marker maps; Genetics 157:1819-1829
2.Heffner EL, Sorrells ME, Jannink JL (2009) Genomic Selection for Crop Improvement. Crop
Science 49:1-12
Download