Genomic Selection Using Pedigree and Marker-by-Environment Interaction for Barley Seed Quality Traits From Two Commercial Breeding Programs

Abstract

With the current advances in the development of low-cost high-density array-based DNA marker technologies, cereal breeding programs are increasingly relying on genomic selection as a tool to accelerate the rate of genetic gain in seed quality traits. Different sources of genetic information are being explored, with the most prevalent being combined additive information from marker and pedigree-based data, and their interaction with the environment. In this, there has been mixed evidence on the performance of use of these data. This study undertook an extensive analysis of 907 elite winter barley (Hordeum vulgare L.) lines across multiple environments from two breeding companies. Six genomic prediction models were evaluated to demonstrate the effect of using pedigree and marker information individually and in combination, as well their interactions with the environment. Each model was evaluated using three cross-validation schemes that allows the prediction of newly developed lines (lines that have not been evaluated in any environment), prediction of new or unobserved years, and prediction of newly developed lines in unobserved years. The results showed that the best prediction model depends on the cross-validation scheme employed. In predicting newly developed lines in known environments, marker information had no advantage to pedigree information. Predictions in this scenario also benefited from including genotype-by-environment interaction in the models. However, when predicting lines and years not observed previously, marker information was superior to pedigree data. Nonetheless, such scenarios did not benefit from the addition of genotype-by-environment interaction. A combination of pedigree-based and markerbased information produced a similar or only marginal improvement in prediction ability. It was also discovered that combining populations from the different breeding programs to increase training population size had no advantage in prediction.

Optimizing predictions in IRRI’s rice drought breeding program by leveraging 17 years of historical data and pedigree information

“…This gain is not surprising and may be due to the significant contribution of variance components contributed by the interaction effects. Thus, we argue that reasonable gains in the prediction ability can be achieved by jointly considering the main and environmental interaction components as depicted in previous studies on wheat (Burgueño et al, 2012;Sukumaran et al, 2017) and barley (Ankamah-Yeboah et al, 2020).…”

“…In general, the incorporation of genotype-byenvironment interactions (pedigree-by-environment; A × E here) in prediction models increased the predictive ability in all three scenarios. Dissecting and quantifying A × E has been an integral part of predictions and has led to increased prediction ability (Jarquín et al, 2017;Ankamah-Yeboah et al, 2020), as found in this study. Further models, involving the GCA and SCA effects along with the A × E interactions have slightly better predictive ability as compared to those only involving the GCA and SCA effects.…”

Multi-Year Dynamics of Single-Step Genomic Prediction in an Applied Wheat Breeding Program

“…Notwithstanding, it should be mentioned that the pedigree-assisted models including phenotypic observations from preliminary yield trials performed sometimes similar or even superior to the corresponding genomic-assisted selection model. This observation was likewise made by the authors of [15,17] when predicting already tested lines across a multitude of international mega-environments, and might be attributed to the ability to differentiate between lines within families if preexisting information of the selection candidates is exploited for making predictions. Further possible influencing factors include the altering of performance rankings by inducing a shrinkage towards family means when modelling covariances between family members [48] as well as the previous mentioned drawing of performance information from half-sibs and full-sibs [49,50].…”

A New Deep Learning Calibration Method Enhances Genome-Based Prediction of Continuous Crop Traits

“…Two reasonable explanations as to why the GBLUP method outperformed the proposed deep learning method even with the proposed calibration method could be that the four data sets are: 1) small and, as pointed out above, the deep learning methods are data hungry, and 2) they do not have complex nonlinear patterns. It is also important to highlight that our results only used markers (not pedigree) information, whereas some researchers have reported similar results using both pedigree and markers (Ankamah-Yeboah et al, 2020;Calleja-Rodriguez, et al, 2020). Furthermore, the training process with the two deep learning methods (DL_M1 and DL_M2) was considerable slower than the conventional GBLUP method.…”