Optimization of genomic selection training populations with a genetic algorithm

Akdemir, Deniz; Sánchez, Julio Isidro; Jannink, Jean‐Luc

doi:10.1186/s12711-015-0116-6

Cited by 138 publications

(184 citation statements)

References 28 publications

Supporting

Mentioning

175

Contrasting

Order By: Relevance

“…Isidro et al (2015) found that for the rice dataset, an approach to developing a training population based on a stratified sampling strategy similar to our approach in the genotypic core gave the highest predictive accuracy compared with training populations chosen by different optimization algorithms. Similarly, Crossa et al (2016) found that prediction accuracy was similar for all traits using a diversity core set and one chosen to minimize prediction error variance when using a method outlined by Akdemir et al (2015). This suggests that stratification based on diversity is a straightforward and effective method to generate subsets of germplasm to be used in genomic prediction.…”

Section: Discussionmentioning

confidence: 92%

Comparison of Representative and Custom Methods of Generating Core Subsets of a Carrot Germplasm Collection

et al. 2019

View full text Add to dashboard Cite

Crop breeding programs are interested in using genetic resources but have difficulty identifying useful accessions from germplasm collections. To efficiently use the diversity present in large germplasm collections, breeders often identify a subset of accessions that represent the larger collection. Methods to identify these subsets, which are called core collections, do not consistently capture functional diversity, and breeders would benefit from methods that help create custom core collections using existing data from variety trials or breeding programs. Making use of high‐density genomic data and existing phenotypic data from a collection of 433 domesticated carrot (Daucus carota L.) accessions, we tested whether it is possible to develop custom subsets of accessions for specific breeding purposes. We found that for this collection, representative strategies were effective in developing core collections that capture the diversity of the collection, but they were no better than random sampling, likely because the collection itself is not strongly subdivided. Custom strategies generated subsets that differed from the total collection with altered genetic, geographic, and phenotypic compositions. When used as training populations for genomic prediction of the other accessions in the collection, however, these custom cores did not produce a substantial improvement over traditional core collections. Increasing the size of the core did improve prediction accuracy, suggesting that it is possible to improve the usefulness of core collections by identifying custom subsets that are large enough to represent the functional genetic diversity present in the collection.

show abstract

Section: Discussionmentioning

confidence: 92%

Comparison of Representative and Custom Methods of Generating Core Subsets of a Carrot Germplasm Collection

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Panicle architecture could be used in indirect or trait‐assisted genomic prediction for GY by breeding programs dominated by durra‐type sorghum varieties. In addition to utilization of within‐ and across‐group genetic variances, optimization algorithms could also help in efficient design of training population for diversity panels and breeding populations (Akdemir, Sanchez, & Jannink, 2015; Isidro et al., 2015).…”

Section: Discussionmentioning

confidence: 99%

Impact of sorghum racial structure and diversity on genomic prediction of grain yield components

et al. 2020

View full text Add to dashboard Cite

Population structure is an important factor that affects the accuracy of estimated breeding values in genomic prediction. Natural sorghum [Sorghum bicolor (L.) Moench] populations exhibit population structure resulting from genetic and morphological differentiation due to evolutionary divergence. To study the impact of sorghum racial structure and diversity in genomic prediction, we conducted two cross‐validation (CV) experiments: CV1, proportional sampling from races; and CV2, sampling from across race (AR) or within race (WR). A diversity panel with 389 individuals with 224,007 single nucleotide polymorphisms was used for genomic prediction. Genomic heritabilities for traits were positively correlated (0.63) with their mean prediction accuracy (r) from CV1, and within‐subpopulation variance accounted for ∼80% of total genetic variance. The CV1 prediction accuracy ranged from 0.52–0.69, but r declined by 39 and 54% on average for WR and AR methods, respectively. As a predictor, race explained 30–50% of covariance for grain and panicle traits, but race was a bad predictor of plant height, as expected. Grain weight was consistently the best predicted trait across CV1 and CV2 methods except in AR. Difference in average r for WR and AR was greater in durra and caudatum, small in kafir, and nonexistent in guinea and mixed subgroups. We observed higher prevalence of minor alleles among guinea and mixed subgroups, highlighting contribution of allelic diversity towards prediction accuracy. Genomic prediction in sorghum will benefit from utilization of interracial diversity, and we emphasize the need for further investigations into the role of racial structure in genomic prediction.

show abstract

“…The design of training populations for GS models attained a lot of attention recently (Rincent et al, 2012; Akdemir et al, 2015; Isidro et al, 2015). This approach is promising because any gain in accuracy or any reduction in experimental costs that can be obtained by carefully designing the training populations will proportionately be realized as gains.…”

Section: Discussionmentioning

confidence: 99%

Efficient Breeding by Genomic Mating

2016

Self Cite

View full text Add to dashboard Cite

Selection in breeding programs can be done by using phenotypes (phenotypic selection), pedigree relationship (breeding value selection) or molecular markers (marker assisted selection or genomic selection). All these methods are based on truncation selection, focusing on the best performance of parents before mating. In this article we proposed an approach to breeding, named genomic mating, which focuses on mating instead of truncation selection. Genomic mating uses information in a similar fashion to genomic selection but includes information on complementation of parents to be mated. Following the efficiency frontier surface, genomic mating uses concepts of estimated breeding values, risk (usefulness) and coefficient of ancestry to optimize mating between parents. We used a genetic algorithm to find solutions to this optimization problem and the results from our simulations comparing genomic selection, phenotypic selection and the mating approach indicate that current approach for breeding complex traits is more favorable than phenotypic and genomic selection. Genomic mating is similar to genomic selection in terms of estimating marker effects, but in genomic mating the genetic information and the estimated marker effects are used to decide which genotypes should be crossed to obtain the next breeding population.

show abstract

Optimization of genomic selection training populations with a genetic algorithm

Cited by 138 publications

References 28 publications

Comparison of Representative and Custom Methods of Generating Core Subsets of a Carrot Germplasm Collection

Comparison of Representative and Custom Methods of Generating Core Subsets of a Carrot Germplasm Collection

Impact of sorghum racial structure and diversity on genomic prediction of grain yield components

Efficient Breeding by Genomic Mating

Contact Info

Product

Resources

About