Genomic selection in plant breeding: from theory to practice

We look forward and consider research needs surrounding methodological questions and the implications of genomic selection for long-term selection

J.-L. Jannink


Scholarcy highlights

  • It has been predicted for over two decades that molecular marker technology would reshape breeding programs and facilitate rapid gains from selection
  • While marker-assisted selection has been effective for the manipulation of large effect alleles with known association to a marker, it has been at an impasse when many alleles of small effect segregate and no substantial, reliable effects can be identified
  • QTL identification methods can make MAS poorly suited to crop improvement: biparental populations may be used that are not representative and in any event do not have the same level of allelic diversity and phase as the breeding program as a whole ; the necessity of generating such populations is costly such that the populations may be small and underpowered; validation of discoveries is warranted, requiring additional effort; the separation of QTL identification from estimation means that estimated effects will be biased, and small-effect QTL will be missed entirely as a result of using stringent significance thresholds
  • Meuwissen et al and Habier et al evaluated the accuracies of ridge regression and BayesB using similar approaches assuming additive gene action and a heritability of 0.5
  • The greater overall accuracy and greater difference between ridge regression and BayesB in Meuwisen et al can probably be attributed to the larger variances generated by individual QTL in that study
  • Rather than seeking to identify individual loci significantly associated with a trait, genomic selection uses all marker data as predictors of performance and delivers more accurate predictions
  • Factors that we looked at that reduced that proportion were fewer QTL, higher marker density, larger training population size, and as expected, BayesB versus ridge regression
  • Existing empirical studies make it clear that the underlying genetic architecture, as characterized at least by the number of QTL and the distributions of their allelic effects and frequencies, differentially affects performance of different genomic selection methods

Need more features? Save interactive summary cards to your Scholarcy Library.