Esa A. Mäntysaari

Esa A. Mäntysaari

Auto-generated profile, based on publicly available information

Esa A. Mäntysaari

Natural Resources Institute Finland (Luke), 31600 Jokioinen, Finland

Research Focus & Metrics

Academic Career Length

20Years (2007-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 November 2017 in Journal of Animal Science
Verified
Journal Article
Closed access
PDF file
EME. A. Mäntysaari
E. A. Mäntysaari
RER. D. Evans
R. D. Evans
ISIsmo Stranden
Ismo Stranden
An equivalent computational approach called ssGTBLUP was formulated for the original single-step GBLUP (ssGBLUP). In ssGTBLUP, the genomic relationship matrix has the form G = ZZʹ + C, where the (centered and scaled) Z marker matrix has size n x m (numbers of genotypes and markers), and the C matrix can be easily inverted. The inverse can be written as G-1 = C-1 − TʹT where T is an m by n matrix. When the preconditioned conjugate gradient (PCG) method is used to solve the mixed model equations, a matrix vector product G-1d needs to be computed. In ssGBLUP, this requires n2 multiplications, but in ssGTBLUP, the product TʹTd has 2nm multiplications and C-1d has cn multiplications with the constant c independent of n or m. In an approximate approach called ssGTBLUP(p), the eigendecomposition of ZʹC-1Z is used to reduce the number of rows in the T matrix. Here, p is the percentage of total variance explained by the accepted eigenvalues. The objective of this study was to compare the performance of ssGBLUP, ssGTBLUP, ssGTBLUP(p), and the APY (algorithm for proven and young) method. In APY, the core had 50,000 (APY50K), 30,000 (APY30K), or 10,000 (APY10K) animals. The approaches were tested on the Irish beef carcass conformation genetic evaluation which has a heterogeneous multibreed population. The pedigree had 13.3 million animals. There were m = 54,620 markers available from n = 163,277 genotyped animals. For genotyped animals, the correlations of breeding values between ssGBLUP and ssGTBLUP(p) for the 11 traits in the model ranged from 0.999–1.000 for p = 99, 0.998–1.000 for p = 98, and 0.992–0.998 for p = 95 but were 0.994–1.000 for APY50K, 0.969–0.997 for APY30K, and 0.899–0.967 for APY10K. Computing times per iteration were 4.43, 3.30, 2.69, 2.29, 1.55, 1.76, 1.27, and 0.55 min for ssGBLUP, ssGTBLUP, ssGTBLUP(99), ssGTBLUP(98), ssGTBLUP(95), APY50K, APY30K, and APY10K, respectively. The ssGTBLUP(p) approach allowed a well-defined approximation to ssGBLUP and fast computations Copyright © 2017. American Society of Animal Science .

1 June 2020 in Journal of Dairy Science
Verified
Journal Article
Closed access
EME.A. Mäntysaari
E.A. Mäntysaari
MKMinna Koivula
Minna Koivula
ISIsmo Stranden
Ismo Stranden
During the last decade, genomic selection has revolutionized dairy cattle breeding. For example, Nordic dairy cows (Denmark, Finland, and Sweden) born in 2018 were >90% sired by young genomically tested bulls. Thus, the average age of sires for Red Dairy Cattle cows born in 2018 was only 3.1 yr, whereas in 2011 it was 5.7 yr. Earlier the key driver of genetic progress was the selection of progeny-tested sires, but now it is the genomic preselection of young sires. This leads to a biased estimation of genetic progress by the traditional genetic evaluations. When these are used as input for multi-step genomic evaluations also they became distorted. The only long-term solution to maintain unbiasedness is to include the genomic information in evaluations. Although means for single-step evaluation models were introduced in 2010, they have not yet been implemented in large-scale national dairy evaluations. At first, single-step evaluations were hindered by computational cost. This has been largely solved, either by sparse presentations of the inverses of the genomic relationship (G) and pedigree relationship (A22) matrices of genotyped animals needed in the single-step evaluation models based on G (ssGBLUP), or by using the single-step marker models. Approaches for G-1 are the APY-G, where the relationships among "young" animals are completely determined by their relationship to the "core" animals, and single-step evaluations where the G-1 is replaced by a computational formula based on the structure of G (ssGTBLUP). The single-step marker models include the marker effects either directly, as effects in the statistical model, or indirectly, to generate genomic relationships among genotyped animals. Concurrently with development of the algorithm, computing resources have evolved in both availability of computer memory and speed. The problems actively studied now are the same for both of the single-step approaches (GBLUP and marker models). Convergence in iterative solving seems to get worse with an increasing number of genotypes. These problems are more pronounced with low-heritability traits and in multi-trait models with high genetic correlations among traits. Problems are also related to the unbalancedness of pedigrees and diverse genetic groups. In many cases, the problem can be solved by properly accounting for contributions of the genotyped animals to genetic groups. The standard solving approach is preconditioned conjugate gradient iteration, in which the convergence has been improved by better preconditioning matrices. Another difficulty to be considered is inflation in genomic evaluations of candidate animals; genomic models seem to overvalue the genomic information. The problem is usually smaller in single-step evaluations than in multi-step evaluations but is more difficult to mitigate by ad hoc adjustments.

30 March 2017 in Genetics Selection Evolution
Verified
Journal Article
Open access
PDF file
MTMatti Taskinen
Matti Taskinen
EMEsa A. Mäntysaari
Esa A. Mäntysaari
ISIsmo Stranden
Ismo Stranden
Single-step genomic best linear unbiased prediction (BLUP) evaluation combines relationship information from pedigree and genomic marker data. The inclusion of the genomic information into mixed model equations requires the inverse of the combined relationship matrix $${\mathbf{H}}$$ H , which has a dense matrix block for genotyped animals. To avoid inversion of dense matrices, single-step genomic BLUP can be transformed to single-step single nucleotide polymorphism BLUP (SNP-BLUP) which have observed and imputed marker coefficients. Simple block LDL type decompositions of the single-step relationship matrix $${\mathbf{H}}$$ H were derived to obtain different types of linearly equivalent single-step genomic mixed model equations with different sets of reparametrized random effects. For non-genotyped animals, the imputed marker coefficient terms in the single-step SNP-BLUP were calculated on-the-fly during the iterative solution using sparse matrix decompositions without storing the imputed genotypes. Residual polygenic effects were added to genotyped animals and transmitted to non-genotyped animals using relationship coefficients that are similar to imputed genotypes. The relationships were further orthogonalized to improve convergence of iterative methods. All presented single-step SNP-BLUP models can be solved efficiently using iterative methods that rely on iteration on data and sparse matrix approaches. The efficiency, accuracy and iteration convergence of the derived mixed model equations were tested with a small dataset that included 73,579 animals of which 2885 were genotyped with 37,526 SNPs. Inversion of the large and dense genomic relationship matrix was avoided in single-step evaluation by using fully orthogonalized single-step SNP-BLUP formulations. The number of iterations until convergence was smaller in single-step SNP-BLUP formulations than in the original single-step GBLUP when heritability was low, but increased above that of the original single-step when heritability was high.

1 July 2020 in Journal of Dairy Science
Verified
Journal Article
Closed access
AKA.A. Kudinov
A.A. Kudinov
EME.A. Mäntysaari
E.A. Mäntysaari
GAGert Pedersen Aamand
Gert Pedersen Aamand
PUP. Uimari
P. Uimari
ISIsmo Stranden
Ismo Stranden
Single-step genomic BLUP (ssGBLUP) is a powerful approach for breeding value prediction in populations with a limited number of genotyped animals. However, conflicting genomic (G) and pedigree (A22) relationship matrices complicate the implementation of ssGBLUP into practice. The metafounder (MF) approach is a recently proposed solution for this problem and has been successfully used on simulated and real multi-breed pig data. Advantages of the method are easily seen across breed evaluations, where pedigrees are traced to several pure breeds, which are thereafter used as MF. Application of the MF method to ruminants is complicated due to multi-breed pedigree structures and the inability to transmit existing unknown parent groups (UPG) to MF. In this study, we apply the MF approach for ssGBLUP evaluation of Finnish Red Dairy cattle treated as a single breed. Relationships among MF were accounted for by a (co)variance matrix (Γ) computed using estimated base population allele frequencies. The attained Γ was used to calculate a relationship matrix A22Γ for the genotyped animals. We tested the influence of SNP selection on the Γ matrix by applying a minor allele frequency (MAF) threshold (ΓMAF) where accepted markers had an MAF ≥0.05. Elements in the ΓMAF matrix were slightly lower than in the Γ matrix. Correlation between diagonal elements of the genomic and pedigree relationship matrices increased from 0.53 (A22) to 0.76 ( A22Γ and [Formula: see text] ). Average diagonal elements of A22Γ and [Formula: see text] matrices increased to the same level as in the G matrix. The ssGBLUP breeding values (GEBV) were solved using either the original 236 or redefined 8 UPG, or 8 MF computed with or without the MAF threshold. For bulls, the GEBV validation test results for the 8 UPG and 8 MF gave the same validation reliability (R2; 0.31) and over-dispersion (0.73, measured by regression coefficient b1). No significant R2 increase was observed in cows. Thus, the MF greatly influenced the pedigree relationship matrices but not the GEBV. Selection of SNP according to MAF had a notable effect on the Γ matrix and made the A22 and G matrices more similar.

16 August 2018 in Journal of Animal Breeding and Genetics
Verified
Journal Article
Closed access
PDF file
KMKaarina Matilainen
Kaarina Matilainen
ISIsmo Stranden
Ismo Stranden
GAGert Pedersen Aamand
Gert Pedersen Aamand
EMEsa A. Mäntysaari
Esa A. Mäntysaari

8 June 2023 in Genetics Selection Evolution
Verified
Journal Article
Open access
PDF file
JVJeremie Vandenplas
Jeremie Vandenplas
JNJan Ten Napel
Jan Ten Napel
SDSaeid Naderi Darbaghshahi
Saeid Naderi Darbaghshahi
RERoss Evans
Ross Evans
MCMario Pl L. Calus
Mario Pl L. Calus
ISIsmo Stranden
Ismo Stranden
Background: Single-step genomic best linear unbiased prediction (ssGBLUP) models allow the combination of genomic, pedigree, and phenotypic data into a single model, which is computationally challenging for large genotyped populations. In practice, genotypes of animals without their own phenotype and progeny, so-called genotyped selection candidates, can become available after genomic breeding values have been estimated by ssGBLUP. In some breeding programmes, genomic estimated breeding values (GEBV) for these animals should be known shortly after obtaining genotype information but recomputing GEBV using the full ssGBLUP takes too much time. In this study, first we compare two equivalent formulations of ssGBLUP models, i.e. one that is based on the Woodbury matrix identity applied to the inverse of the genomic relationship matrix, and one that is based on marker equations. Second, we present computationally-fast approaches to indirectly compute GEBV for genotyped selection candidates, without the need to do the full ssGBLUP evaluation. Results: The indirect approaches use information from the latest ssGBLUP evaluation and rely on the decomposition of GEBV into its components. The two equivalent ssGBLUP models and indirect approaches were tested on a six-trait calving difficulty model using Irish dairy and beef cattle data that include 2.6 million genotyped animals of which about 500,000 were considered as genotyped selection candidates. When using the same computational approaches, the solving phase of the two equivalent ssGBLUP models showed similar requirements for memory and time per iteration. The computational differences between them were due to the preprocessing phase of the genomic information. Regarding the indirect approaches, compared to GEBV obtained from single-step evaluations including all genotypes, indirect GEBV had correlations higher than 0.99 for all traits while showing little dispersion and level bias. Conclusions: In conclusion, ssGBLUP predictions for the genotyped selection candidates were accurately approximated using the presented indirect approaches, which are more memory efficient and computationally fast, compared to solving a full ssGBLUP evaluation. Thus, indirect approaches can be used even on a weekly basis to estimate GEBV for newly genotyped animals, while the full single-step evaluation is done only a few times within a year.