13 Intro to Genomic Cross Prediction

Genomic prediction of the mean, variance and usefulness of crosses can be accomplished with genomicMateSelectR functions.

13.1 Understanding mate selection

Click Here for a google slides presentation entitled “Genomic mate selection in outbred species: predicting cross usefulness with additive and total genetic covariance matrices” where I introduce the concepts.

The theory / formulae are summarized as part of this genomicMateSelectR vignette

13.1.1 Literature

Here are some recommended articles to read regarding genomic mate selection (full citations will be at bottom): (Bonk et al. 2016; Lehermeier et al. 2017; Neyhart and Smith 2019; Neyhart et al. 2019; Bijma et al. 2020; Werner et al. 2020; Wolfe et al. 2021)

In particular, read these:

1. Wolfe et a. 2021.Genomic mating in outbred species: predicting cross usefulness with additive and total genetic covariance matrices. https://doi.org/10.1093/genetics/iyab122.
2. Werner et al. 2020. Genomic selection strategies for clonally propagated crops. https://doi.org/10.1101/2020.06.15.152017.

You may very well need/want to read the literature that are referenced in these articles. If you do that, you’ll have a solid foundation for understanding prediction of cross performance.

13.1.2 Tutorial

For an a tutorial on how to execute these predictions using genomicMateSelectR functions, see this “Getting started predicting crosses vignette”.

13.2 Non-additive effects

Up until now, we have used an additive-effects only model (modelType="A"), which gives us access to predictions of GEBV.

In addition, genomicMateSelectR enables two types of non-additive effects models to be implemented: an additive plus dominance model (modelType="AD") and a directional dominance model that allows for an inbreeding depression (or heterotic) effect (modelType="DirDom").

13.2.1 Literature

• For the basic quantitative genetics concepts of additive and dominance effects:

  • Intro to Quantitative Genetics, part of Felipe Ferrão’s “Survey of Breeding Tools and (Genomic Selection) Methods.”
  • See also the list of Recommended Literature provided in the previous chapter Intro to Genomic Prediction

• See my summary of the additive and non-additive “genomic prediction models implemented” as part of the 2nd genomicMateSelectR vignette. See also the references to the literature, which are cited there.

• Two good papers to start studying genomic prediction with non-additive effects are:

  • Vitezica et al. 2013. “On the Additive and Dominant Variance and Covariance of Individuals Within the Genomic Selection Scope.” Genetics 195 (4): 1223–30. https://doi.org/10.1534/genetics.113.155176

  • Varona et al. 2013. “Non-Additive Effects in Genomic Selection.” Frontiers in Genetics 9 (March). https://doi.org/10.3389/fgene.2018.00078

13.2.2 Tutorial

The vignette in genomicMateSelectR entitled “Genomic prediction with non-additive effects”provides a complete tutorial on how to execute these models and predicting cross-performance with them.

13.3 Parent-wise Cross-validation

How can we estimate the accuracy of predicting previously untested crosses?

In the mate selection article, Wolfe et al. (2021) I devised a cross-validation strategy that uses a pedigree -based approach, which I called “parent-wise cross-validation.” The approach is described in detail in the manuscript. It is illustrated starting on Slide 50 of this gSlides presentation.

genomicMateSelectR provides a function runParentWiseCrossVal() (see here for the documentation / details) to implement this kind of cross-validation.

An example of it’s implementation in-practice is part of the IITA 2021 Genomic Selection documentation here.

In the next section, I will attempt a smaller example using the data we have been working with in this manual.