Grouping of genomic markers in populations with family structure.

BACKGROUND: Linkage and linkage disequilibrium (LD) between genome regions cause dependencies among genomic markers. Due to family stratification in populations with non-random mating in livestock or crop, the standard measures of population LD such as [Formula: see text] may be biased. Grouping of markers according to their interdependence needs to account for the actual population structure in order to allow proper inference in genome-based evaluations. RESULTS: Given a matrix reflecting the strength of association between markers, groups are built successively using a greedy algorithm; largest groups are built at first. As an option, a representative marker is selected for each group. We provide an implementation of the grouping approach as a new function to the R package hscovar. This package enables the calculation of the theoretical covariance between biallelic markers for half- or full-sib families and the derivation of representative markers. In case studies, we have shown that the number of groups comprising dependent markers was smaller and representative SNPs were spread more uniformly over the investigated chromosome region when the family stratification was respected compared to a population-LD approach. In a simulation study, we observed that sensitivity and specificity of a genome-based association study improved if selection of representative markers took family structure into account. CONCLUSIONS: Chromosome segments which frequently recombine in the underlying population can be identified from the matrix of pairwise dependence between markers. Representative markers can be exploited, for instance, for dimension reduction prior to a genome-based association study or the grouping structure itself can be employed in a grouped penalization approach.

Seagull: lasso, group lasso and sparse-group lasso regularization for linear regression models via proximal gradient descent.

BACKGROUND: Statistical analyses of biological problems in life sciences often lead to high-dimensional linear models. To solve the corresponding system of equations, penalization approaches are often the methods of choice. They are especially useful in case of multicollinearity, which appears if the number of explanatory variables exceeds the number of observations or for some biological reason. Then, the model goodness of fit is penalized by some suitable function of interest. Prominent examples are the lasso, group lasso and sparse-group lasso. Here, we offer a fast and numerically cheap implementation of these operators via proximal gradient descent. The grid search for the penalty parameter is realized by warm starts. The step size between consecutive iterations is determined with backtracking line search. Finally, seagull -the R package presented here- produces complete regularization paths. RESULTS: Publicly available high-dimensional methylation data are used to compare seagull to the established R package SGL. The results of both packages enabled a precise prediction of biological age from DNA methylation status. But even though the results of seagull and SGL were very similar (R2 > 0.99), seagull computed the solution in a fraction of the time needed by SGL. Additionally, seagull enables the incorporation of weights for each penalized feature. CONCLUSIONS: The following operators for linear regression models are available in seagull: lasso, group lasso, sparse-group lasso and Integrative LASSO with Penalty Factors (IPF-lasso). Thus, seagull is a convenient envelope of lasso variants.

A Fitted Sparse-Group Lasso for Genome-Based Evaluations.

In life sciences, high-throughput techniques typically lead to high-dimensional data and often the number of covariates is much larger than the number of observations. This inherently comes with multicollinearity challenging a statistical analysis in a linear regression framework. Penalization methods such as the lasso, ridge regression, the group lasso, and convex combinations thereof, which introduce additional conditions on regression variables, have proven themselves effective. In this study, we introduce a novel approach by combining the lasso and the standardized group lasso leading to meaningful weighting of the predicted ("fitted") outcome which is of primary importance, e.g., in breeding populations. This "fitted" sparse-group lasso was implemented as a proximal-averaged gradient descent method and is part of the R package "seagull" available at CRAN. For the evaluation of the novel method, we executed an extensive simulation study. We simulated genotypes and phenotypes which resemble data of a dairy cattle population. Genotypes at thousands of genomic markers were used as covariates to fit a quantitative response. The proximity of markers on a chromosome determined grouping. In the majority of simulated scenarios, the new method revealed improved prediction abilities compared to other penalization approaches and was able to localize the signals of simulated features.

An ultra-high density SNP-based linkage map for enhancing the pikeperch (Sander lucioperca) genome assembly to chromosome-scale.

Pikeperch (Sander lucioperca) is a fish species with growing economic significance in the aquaculture industry. However, successful positioning of pikeperch in large-scale aquaculture requires advances in our understanding of its genome organization. In this study, an ultra-high density linkage map for pikeperch comprising 24 linkage groups and 1,023,625 single nucleotide polymorphisms markers was constructed after genotyping whole-genome sequencing data from 11 broodstock and 363 progeny, belonging to 6 full-sib families. The sex-specific linkage maps spanned a total of 2985.16 cM in females and 2540.47 cM in males with an average inter-marker distance of 0.0030 and 0.0026 cM, respectively. The sex-averaged map spanned a total of 2725.53 cM with an average inter-marker distance of 0.0028 cM. Furthermore, the sex-averaged map was used for improving the contiguity and accuracy of the current pikeperch genome assembly. Based on 723,360 markers, 706 contigs were anchored and oriented into 24 pseudomolecules, covering a total of 896.48 Mb and accounting for 99.47% of the assembled genome size. The overall contiguity of the assembly improved with a scaffold N50 length of 41.06 Mb. Finally, an updated annotation of protein-coding genes and repetitive elements of the enhanced genome assembly is provided at NCBI.

Covariance Between Genotypic Effects and its Use for Genomic Inference in Half-Sib Families.

In livestock, current statistical approaches utilize extensive molecular data, e.g., single nucleotide polymorphisms (SNPs), to improve the genetic evaluation of individuals. The number of model parameters increases with the number of SNPs, so the multicollinearity between covariates can affect the results obtained using whole genome regression methods. In this study, dependencies between SNPs due to linkage and linkage disequilibrium among the chromosome segments were explicitly considered in methods used to estimate the effects of SNPs. The population structure affects the extent of such dependencies, so the covariance among SNP genotypes was derived for half-sib families, which are typical in livestock populations. Conditional on the SNP haplotypes of the common parent (sire), the theoretical covariance was determined using the haplotype frequencies of the population from which the individual parent (dam) was derived. The resulting covariance matrix was included in a statistical model for a trait of interest, and this covariance matrix was then used to specify prior assumptions for SNP effects in a Bayesian framework. The approach was applied to one family in simulated scenarios (few and many quantitative trait loci) and using semireal data obtained from dairy cattle to identify genome segments that affect performance traits, as well as to investigate the impact on predictive ability. Compared with a method that does not explicitly consider any of the relationship among predictor variables, the accuracy of genetic value prediction was improved by 10-22%. The results show that the inclusion of dependence is particularly important for genomic inference based on small sample sizes.