We review and propose several methods for identifying possible outliers and evaluate their properties. The methods are applied to a genomic prediction program in hybrid rye. Many plant breeders use ANOVA-based software for routine analysis of field trials. These programs may offer specific in-built options for residual analysis that are lacking in current REML software. With the advance of molecular technologies, there is a need to switch to REML-based approaches, but without losing the good features of outlier detection methods that have proven useful in the past. Our aims were to compare the variance component estimates between ANOVA and REML approaches, to scrutinize the outlier detection method of the ANOVA-based package PlabStat and to propose and evaluate alternative procedures for outlier detection. We compared the outputs produced using ANOVA and REML approaches of four published datasets of generalized lattice designs. Five outlier detection methods are explained step by step. Their performance was evaluated by measuring the true positive rate and the false positive rate in a dataset with artificial outliers simulated in several scenarios. An implementation of genomic prediction using an empirical rye multi-environment trial was used to assess the outlier detection methods with respect to the predictive abilities of a mixed model for each method. We provide a detailed explanation of how the PlabStat outlier detection methodology can be translated to REML-based software together with the evaluation of alternative methods to identify outliers. The method combining the Bonferroni-Holm test to judge each residual and the residual standardization strategy of PlabStat exhibited good ability to detect outliers in small and large datasets and under a genomic prediction application. We recommend the use of outlier detection methods as a decision support in the routine data analyses of plant breeding experiments.

BackgroundGenomic prediction is becoming a daily tool for plant breeders. It makes use of genotypic information to make predictions used for selection decisions. The accuracy of the predictions depends on the number of genotypes used in the calibration; hence, there is a need of combining data across years. A proper phenotypic analysis is a crucial prerequisite for accurate calibration of genomic prediction procedures. We compared stage-wise approaches to analyse a real dataset of a multi-environment trial (MET) in rye, which was connected between years only through one check, and used different spatial models to obtain better estimates, and thus, improved predictive abilities for genomic prediction. The aims of this study were to assess the advantage of using spatial models for the predictive abilities of genomic prediction, to identify suitable procedures to analyse a MET weakly connected across years using different stage-wise approaches, and to explore genomic prediction as a tool for selection of models for phenotypic data analysis.ResultsUsing complex spatial models did not significantly improve the predictive ability of genomic prediction, but using row and column effects yielded the highest predictive abilities of all models. In the case of MET poorly connected between years, analysing each year separately and fitting year as a fixed effect in the genomic prediction stage yielded the most realistic predictive abilities. Predictive abilities can also be used to select models for phenotypic data analysis. The trend of the predictive abilities was not the same as the traditionally used Akaike information criterion, but favoured in the end the same models.ConclusionsMaking predictions using weakly linked datasets is of utmost interest for plant breeders. We provide an example with suggestions on how to handle such cases. Rather than relying on checks we show how to use year means across all entries for integrating data across years. It is further shown that fitting of row and column effects captures most of the heterogeneity in the field trials analysed.Electronic supplementary materialThe online version of this article (doi:10.1186/1471-2164-15-646) contains supplementary material, which is available to authorized users.

Key message Hyperspectral and genomic data are effective predictors of biomass yield in winter rye. Variable selection procedures can improve the informativeness of reflectance data. Abstract Integrating cutting-edge technologies is imperative to sustainably breed crops for a growing global population. To predict dry matter yield (DMY) in winter rye (Secale cereale L.), we tested single-kernel models based on genomic (GBLUP) and hyperspectral reflectance-derived (HBLUP) relationship matrices, a multi-kernel model combining both matrices and a bivariate model fitted with plant height as a secondary trait. In total, 274 elite rye lines were genotyped using a 10 k-SNP array and phenotyped as testcrosses for DMY and plant height at four locations in Germany in two years (eight environments). Spectral data consisted of 400 discrete narrow bands ranging between 410 and 993 nm collected by an unmanned aerial vehicle (UAV) on two dates on each environment. To reduce data dimensionality, variable selection of bands was performed, resulting in the least absolute shrinkage and selection operator (Lasso) as the best method in terms of predictive abilities. The mean heritability of reflectance data was moderate ($$h^{2}$$ h 2  = 0.72) and highly variable across the spectrum. Correlations between DMY and single bands were generally significant (p < 0.05) but low (≤ 0.29). Across environments and training set (TRN) sizes, the bivariate model showed the highest prediction abilities (0.56–0.75), followed by the multi-kernel (0.45–0.71) and single-kernel (0.33–0.61) models. With reduced TRN, HBLUP performed better than GBLUP. The HBLUP model fitted with a set of selected bands was preferred. Within and across environments, prediction abilities increased with larger TRN. Our results suggest that in the era of digital breeding, the integration of high-throughput phenotyping and genomic selection is a promising strategy to achieve superior selection gains in hybrid rye.

BackgroundThe use of multiple genetic backgrounds across years is appealing for genomic prediction (GP) because past years’ data provide valuable information on marker effects. Nonetheless, single-year GP models are less complex and computationally less demanding than multi-year GP models. In devising a suitable analysis strategy for multi-year data, we may exploit the fact that even if there is no replication of genotypes across years, there is plenty of replication at the level of marker loci. Our principal aim was to evaluate different GP approaches to simultaneously model genotype-by-year (GY) effects and breeding values using multi-year data in terms of predictive ability. The models were evaluated under different scenarios reflecting common practice in plant breeding programs, such as different degrees of relatedness between training and validation sets, and using a selected fraction of genotypes in the training set. We used empirical grain yield data of a rye hybrid breeding program. A detailed description of the prediction approaches highlighting the use of kinship for modeling GY is presented.ResultsUsing the kinship to model GY was advantageous in particular for datasets disconnected across years. On average, predictive abilities were 5% higher for models using kinship to model GY over models without kinship. We confirmed that using data from multiple selection stages provides valuable GY information and helps increasing predictive ability. This increase is on average 30% higher when the predicted genotypes are closely related with the genotypes in the training set. A selection of top-yielding genotypes together with the use of kinship to model GY improves the predictive ability in datasets composed of single years of several selection cycles.ConclusionsOur results clearly demonstrate that the use of multi-year data and appropriate modeling is beneficial for GP because it allows dissecting GY effects from genomic estimated breeding values. The model choice, as well as ensuring that the predicted candidates are sufficiently related to the genotypes in the training set, are crucial.Electronic supplementary materialThe online version of this article (doi:10.1186/s12863-017-0512-8) contains supplementary material, which is available to authorized users.

Key message Hyperspectral data is a promising complement to genomic data to predict biomass under scenarios of low genetic relatedness. Sufficient environmental connectivity between data used for model training and validation is required. Abstract The demand for sustainable sources of biomass is increasing worldwide. The early prediction of biomass via indirect selection of dry matter yield (DMY) based on hyperspectral and/or genomic prediction is crucial to affordably untap the potential of winter rye (Secale cereale L.) as a dual-purpose crop. However, this estimation involves multiple genetic backgrounds and genetic relatedness is a crucial factor in genomic selection (GS). To assess the prospect of prediction using reflectance data as a suitable complement to GS for biomass breeding, the influence of trait heritability ($$H^{2}$$ H 2 ) and genetic relatedness were compared. Models were based on genomic (GBLUP) and hyperspectral reflectance-derived (HBLUP) relationship matrices to predict DMY and other biomass-related traits such as dry matter content (DMC) and fresh matter yield (FMY). For this, 270 elite rye lines from nine interconnected bi-parental families were genotyped using a 10 k-SNP array and phenotyped as testcrosses at four locations in two years (eight environments). From 400 discrete narrow bands (410 nm–993 nm) collected by an uncrewed aerial vehicle (UAV) on two dates in each environment, 32 hyperspectral bands previously selected by Lasso were incorporated into a prediction model. HBLUP showed higher prediction abilities (0.41 – 0.61) than GBLUP (0.14 – 0.28) under a decreased genetic relationship, especially for mid-heritable traits (FMY and DMY), suggesting that HBLUP is much less affected by relatedness and $$H^{2}$$ H 2 . However, the predictive power of both models was largely affected by environmental variances. Prediction abilities for DMY were further enhanced (up to 20%) by integrating both matrices and plant height into a bivariate model. Thus, data derived from high-throughput phenotyping emerges as a suitable strategy to efficiently leverage selection gains in biomass rye breeding; however, sufficient environmental connectivity is needed.