Prediction of genetic values has been a focus of applied quantitative genetics since the beginning of the 20th century, with renewed interest following the advent of the era of whole genome-enabled prediction. Opportunities offered by the emergence of high-dimensional genomic data fueled by post-Sanger sequencing technologies, especially molecular markers, have driven researchers to extend Ronald Fisher and Sewall Wright's models to confront new challenges. In particular, kernel methods are gaining consideration as a regression method of choice for genome-enabled prediction. Complex traits are presumably influenced by many genomic regions working in concert with others (clearly so when considering pathways), thus generating interactions. Motivated by this view, a growing number of statistical approaches based on kernels attempt to capture non-additive effects, either parametrically or non-parametrically. This review centers on whole-genome regression using kernel methods applied to a wide range of quantitative traits of agricultural importance in animals and plants. We discuss various kernel-based approaches tailored to capturing total genetic variation, with the aim of arriving at an enhanced predictive performance in the light of available genome annotation information. Connections between prediction machines born in animal breeding, statistics, and machine learning are revisited, and their empirical prediction performance is discussed. Overall, while some encouraging results have been obtained with non-parametric kernels, recovering non-additive genetic variation in a validation dataset remains a challenge in quantitative genetics.

Precision animal agriculture is poised to rise to prominence in the livestock enterprise in the domains of management, production, welfare, sustainability, health surveillance, and environmental footprint. Considerable progress has been made in the use of tools to routinely monitor and collect information from animals and farms in a less laborious manner than before. These efforts have enabled the animal sciences to embark on information technology-driven discoveries to improve animal agriculture. However, the growing amount and complexity of data generated by fully automated, high-throughput data recording or phenotyping platforms, including digital images, sensor and sound data, unmanned systems, and information obtained from real-time noninvasive computer vision, pose challenges to the successful implementation of precision animal agriculture. The emerging fields of machine learning and data mining are expected to be instrumental in helping meet the daunting challenges facing global agriculture. Yet, their impact and potential in “big data” analysis have not been adequately appreciated in the animal science community, where this recognition has remained only fragmentary. To address such knowledge gaps, this article outlines a framework for machine learning and data mining and offers a glimpse into how they can be applied to solve pressing problems in animal sciences.

The accessibility of high‐throughput phenotyping platforms in both the greenhouse and field, as well as the relatively low cost of unmanned aerial vehicles, has provided researchers with an effective means to characterize large populations throughout the growing season. These longitudinal phenotypes can provide important insight into plant development and responses to the environment. Despite the growing use of these new phenotyping approaches in plant breeding, the use of genomic prediction models for longitudinal phenotypes is limited in major crop species. The objective of this study was to demonstrate the utility of random regression ( RR ) models using Legendre polynomials for genomic prediction of shoot growth trajectories in rice ( Oryza sativa ). An estimate of shoot biomass, projected shoot area ( PSA ), was recorded over a period of 20 days for a panel of 357 diverse rice accessions using an image‐based greenhouse phenotyping platform. A RR that included a fixed second‐order Legendre polynomial, a random second‐order Legendre polynomial for the additive genetic effect, a first‐order Legendre polynomial for the environmental effect, and heterogeneous residual variances was used to model PSA trajectories. The utility of the RR model over a single time point ( TP ) approach, where PSA is fit at each time point independently, is shown through four prediction scenarios. In the first scenario, the RR and TP approaches were used to predict PSA for a set of lines lacking phenotypic data. The RR approach showed a 11.6% increase in prediction accuracy over the TP approach. Much of this improvement could be attributed to the greater additive genetic variance captured by the RR approach. The remaining scenarios focused forecasting future phenotypes using a subset of early time points for known lines with phenotypic data, as well new lines lacking phenotypic data. In all cases, PSA could be predicted with high accuracy ( r : 0.79 to 0.89 and 0.55 to 0.58 for known and unknown lines, respectively). This study provides the first application of RR models for genomic prediction of a longitudinal trait in rice and demonstrates that RR models can be effectively used to improve the accuracy of genomic prediction for complex traits compared to a TP approach.

Recent work has suggested that the performance of prediction models for complex traits may depend on the architecture of the target traits. Here we compared several prediction models with respect to their ability of predicting phenotypes under various statistical architectures of gene action: (1) purely additive, (2) additive and dominance, (3) additive, dominance, and two-locus epistasis, and (4) purely epistatic settings. Simulation and a real chicken dataset were used. Fourteen prediction models were compared: BayesA, BayesB, BayesC, Bayesian LASSO, Bayesian ridge regression, elastic net, genomic best linear unbiased prediction, a Gaussian process, LASSO, random forests, reproducing kernel Hilbert spaces regression, ridge regression (best linear unbiased prediction), relevance vector machines, and support vector machines. When the trait was under additive gene action, the parametric prediction models outperformed non-parametric ones. Conversely, when the trait was under epistatic gene action, the non-parametric prediction models provided more accurate predictions. Thus, prediction models must be selected according to the most probably underlying architecture of traits. In the chicken dataset examined, most models had similar prediction performance. Our results corroborate the view that there is no universally best prediction models, and that the development of robust prediction models is an important research objective.

The genomic prediction of unobserved genetic values or future phenotypes for complex traits has revolutionized agriculture and human medicine. Fertility traits are undoubtedly complex traits of great economic importance to the dairy industry. Although genomic prediction for improved cow fertility has received much attention, bull fertility largely has been ignored. The first aim of this study was to investigate the feasibility of genomic prediction of sire conception rate (SCR) in US Holstein dairy cattle. Standard genomic prediction often ignores any available information about functional features of the genome, although it is believed that such information can yield more accurate and more persistent predictions. Hence, the second objective was to incorporate prior biological information into predictive models and evaluate their performance. The analyses included the use of kernel-based models fitting either all single nucleotide polymorphisms (SNP; 55K) or only markers with presumed functional roles, such as SNP linked to Gene Ontology or Medical Subject Heading terms related to male fertility, or SNP significantly associated with SCR. Both single- and multikernel models were evaluated using linear and Gaussian kernels. Predictive ability was evaluated in 5-fold cross-validation. The entire set of SNP exhibited predictive correlations around 0.35. Neither Gene Ontology nor Medical Subject Heading gene sets achieved predictive abilities higher than their counterparts using random sets of SNP. Notably, kernel models fitting significant SNP achieved the best performance with increases in accuracy up to 5% compared with the standard whole-genome approach. Models fitting Gaussian kernels outperformed their counterparts fitting linear kernels irrespective of the set of SNP. Overall, our findings suggest that genomic prediction of bull fertility is feasible in dairy cattle. This provides potential for accurate genome-guided decisions, such as early culling of bull calves with low SCR predictions. In addition, exploiting nonlinear effects through the use of Gaussian kernels together with the incorporation of relevant markers seems to be a promising alternative to the standard approach. The inclusion of gene set results into prediction models deserves further research.

BackgroundIn genome-wide studies, over-representation analysis (ORA) against a set of genes is an essential step for biological interpretation. Many gene annotation resources and software platforms for ORA have been proposed. Recently, Medical Subject Headings (MeSH) terms, which are annotations of PubMed documents, have been used for ORA. MeSH enables the extraction of broader meaning from the gene lists and is expected to become an exhaustive annotation resource for ORA. However, the existing MeSH ORA software platforms are still not sufficient for several reasons.ResultsIn this work, we developed an original MeSH ORA framework composed of six types of R packages, including MeSH.db, MeSH.AOR.db, MeSH.PCR.db, the org.MeSH.XXX.db-type packages, MeSHDbi, and meshr.ConclusionsUsing our framework, users can easily conduct MeSH ORA. By utilizing the enriched MeSH terms, related PubMed documents can be retrieved and saved on local machines within this framework.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-015-0453-z) contains supplementary material, which is available to authorized users.

BackgroundArguably, genotypes and phenotypes may be linked in functional forms that are not well addressed by the linear additive models that are standard in quantitative genetics. Therefore, developing statistical learning models for predicting phenotypic values from all available molecular information that are capable of capturing complex genetic network architectures is of great importance. Bayesian kernel ridge regression is a non-parametric prediction model proposed for this purpose. Its essence is to create a spatial distance-based relationship matrix called a kernel. Although the set of all single nucleotide polymorphism genotype configurations on which a model is built is finite, past research has mainly used a Gaussian kernel.ResultsWe sought to investigate the performance of a diffusion kernel, which was specifically developed to model discrete marker inputs, using Holstein cattle and wheat data. This kernel can be viewed as a discretization of the Gaussian kernel. The predictive ability of the diffusion kernel was similar to that of non-spatial distance-based additive genomic relationship kernels in the Holstein data, but outperformed the latter in the wheat data. However, the difference in performance between the diffusion and Gaussian kernels was negligible.ConclusionsIt is concluded that the ability of a diffusion kernel to capture the total genetic variance is not better than that of a Gaussian kernel, at least for these data. Although the diffusion kernel as a choice of basis function may have potential for use in whole-genome prediction, our results imply that embedding genetic markers into a non-Euclidean metric space has very small impact on prediction. Our results suggest that use of the black box Gaussian kernel is justified, given its connection to the diffusion kernel and its similar predictive performance.

Genetic connectedness refers to a measure of genetic relatedness across management units (e.g., herds and flocks). With the presence of high genetic connectedness in management units, best linear unbiased prediction (BLUP) is known to provide reliable comparisons between estimated genetic values. Genetic connectedness has been studied for pedigree-based BLUP; however, relatively little attention has been paid to using genomic information to measure connectedness. In this study, we assessed genome-based connectedness across management units by applying prediction error variance of difference (PEVD), coefficient of determination (CD), and prediction error correlation r to a combination of computer simulation and real data (mice and cattle). We found that genomic information (boldnormalG) increased the estimate of connectedness among individuals from different management units compared to that based on pedigree (boldnormalA). A disconnected design benefited the most. In both datasets, PEVD and CD statistics inferred increased connectedness across units when using boldnormalG- rather than boldnormalA-based relatedness, suggesting stronger connectedness. With r once using allele frequencies equal to one-half or scaling boldnormalG to values between 0 and 2, which is intrinsic to A, connectedness also increased with genomic information. However, PEVD occasionally increased, and r decreased when obtained using the alternative form of G, instead suggesting less connectedness. Such inconsistencies were not found with CD. We contend that genomic relatedness strengthens measures of genetic connectedness across units and has the potential to aid genomic evaluation of livestock species.