Forecasting the accuracy of genomic prediction with different selection targets in the training and prediction set as well as truncation selection
- 560 Downloads
Deterministic formulas accurately forecast the decline in predictive ability of genomic prediction with changing testers, target environments or traits and truncation selection.
Genomic prediction of testcross performance (TP) was found to be a promising selection tool in hybrid breeding as long as the same tester and environments are used in the training and prediction set. In practice, however, selection targets often change in terms of testers, target environments or traits leading to a reduced predictive ability. Hence, it would be desirable to estimate for given training data the expected decline in the predictive ability of genomic prediction under such settings by deterministic formulas that require only quantitative genetic parameters available from the breeding program. Here, we derived formulas for forecasting the predictive ability under different selection targets in the training and prediction set and applied these to predict the TP of lines based on line per se or testcross evaluations. On the basis of two experiments with maize, we validated our approach in four scenarios characterized by different selection targets. Forecasted and empirically observed predictive abilities obtained by cross-validation generally agreed well, with deviations between −0.06 and 0.01 only. Applying the prediction model to a different tester and/or year reduced the predictive ability by not more than 18 %. Accounting additionally for truncation selection in our formulas indicated a substantial reduction in predictive ability in the prediction set, amounting, e.g., to 53 % for a selected fraction α = 10 %. In conclusion, our deterministic formulas enable forecasting the predictive abilities of new selection targets with sufficient precision and could be used to calculate parameters required for optimizing the allocation of resources in multi-stage genomic selection.
KeywordsPrediction Accuracy Predictive Ability Genomic Selection Double Haploid Line Genomic Prediction
The authors thank F. Mauch, J. Jesse, H. Poeschel, R. Lutz, T. Schmidt, S. Pluskat and R. Volkhausen for their assistance in conducting the field experiments. Funding for this research came from the German Federal Ministry of Education and Research (BMBF) within the framework of the projects GABI-Energy (FK 0315045B) and Cornfed (FK 03115461A) as well as from Syngenta under a Ph.D. fellowship for PS.
Compliance with ethical standards
Conflict of interest
The authors declare that they have no conflict of interest.
The authors declare that all experiments comply with the current laws in Germany.
- Cochran WG (1950) Improvement by means of selection. In: Proceedings of second Berkeley symposium on mathematical statistics and probability, University of California Press, pp 449–470Google Scholar
- Dekkers JCM (2007) Prediction of response to marker-assisted and genomic selection using selection index theory. Genetics 124:331–341Google Scholar
- Falconer DS, Mackay TFC (1996) Introduction to quantitative genetics, 4th edn. Longmans Green, HarlowGoogle Scholar
- Foiada F, Westermeier P, Kessel B, Ouzunova M, Wimmer V, Mayerhofer W, Presterl T, Dilger M, Kreps R, Eder J, Schön CC (2015) Improving resistance to the European corn borer: a comprehensive study in elite maize using QTL mapping and genome-wide prediction. Theor Appl Genet 128:875–891CrossRefPubMedGoogle Scholar
- Gilmour A, Gogel B, Cullis B, Thompson R (2009) ASReml user guide release 3.0. VSN International Ltd, Hemel HempsteadGoogle Scholar
- Hallauer AR (1990) Methods used in developing maize inbreds. Maydica 35:1–16Google Scholar
- Utz H (1969) Mehrstufenselektion in der Pflanzenzüchtung. Dissertation Thesis, University of HohenheimGoogle Scholar
- Utz H (2005) PLABSTAT—a computer program for statistical analysis of plant breeding experiments. University of Hohenheim, StuttgartGoogle Scholar