Background:
In the past ten years, genome selection has completely changed dairy cattle breeding. For example, Nordic cattle (Denmark, Finland, Sweden) born at 20 18 >; 90% are bred by young bulls whose genomes have been tested. Therefore, the average bull age of red bull born in 20 18 years is only 3. 1 year, while 20 1 1 year is 5.7 years old. Earlier, the key driving force of genetic progress was the selection of bulls for offspring testing, but now it is the genome pre-selection of young bulls. This leads to the deviation of traditional genetic evaluation in the estimation of genetic progress.
Question:
When these are used as inputs for multi-step genome evaluation, they are also distorted. The only long-term solution to maintain fairness is to include genomic information in the assessment. Although 20 10 introduced the one-step evaluation model, it has not been implemented in the large-scale national dairy industry evaluation. At first, one-step evaluation was hindered by calculating the cost. To a great extent, this has been solved by sparse representation of the inverse matrix of the genome relationship (G) and pedigree relationship (A22) matrices required in the G-based one-step evaluation model (ssGBLUP), or by using the one-step labeling model. The method of G- 1 is APY-G, in which the relationship between "young" animals depends entirely on their relationship with "core" animals, and one-step evaluation, in which G- 1 is replaced by ssGTBLUP based on calculation formula.
One-step marker model includes marker effect, either directly as an effect in statistical model or indirectly producing genomic relationship between genotyping animals. With the development of algorithm, the availability and speed of computing resources in computer memory have been developed. The problem that is actively studied now is the same for both one-step methods (GBLUP and marking model). With the increase of genotype number, the convergence of iterative solution seems to get worse. These problems are more obvious in the multi-trait model with low genetic traits and high genetic correlation among traits.
The problem is also related to unbalanced pedigree and different recombinant inbred lines. In many cases, this problem can be solved by properly considering the contribution of genotypic animals to the genome. The standard solution method is preconditioned yoke gradient iteration, in which the convergence is improved by better preconditioned matrix.
Another difficulty to be considered is the extension of candidate animal genome evaluation; The genome model seems to overestimate the genome information. The problems in single-step evaluation are usually smaller than those in multi-step evaluation, but they are more difficult to alleviate through temporary adjustment.
Summary:
The principle of genome evaluation was put forward more than 10 years ago. At present, genome selection is the main source of genetic improvement of dairy cows in the world. Because this selection cannot be attributed to the selection in cow records or the average value of bull offspring, it will not be captured by genetic evaluation based on pedigree. So EBV began to be more and more biased. At present, bulls born at 20 15 are getting the first batch of offspring test results. These bulls and their bulls were pre-selected according to genome evaluation. In order to ensure the fairness of future assessments, the only option is to start incorporating genomic information into national assessments. This can be done by using one-step evaluation method.
In the past ten years, genotyping has also become a daily tool for choosing dairy cow substitutes. Many countries have also genotyped dairy cows to increase the reference population for genome evaluation. The ultimate goal may be to genotype all animals. At the same time, national genetic assessment must be able to deal with all non-genotypic animals and an increasing number of genotypic animals. When introducing the one-step genome evaluation model, people are worried that the application will have an upper limit on how many genotypes can be included in the evaluation. Using the current one-step technology, the calculation cost is linearly related to the number of animals genotyped; Therefore, the algorithm can deal with any size of dairy cattle population. One-step model with genome relationship can be used ssGTBLUP(M? Ntysaari et al., 20 17) or one-step apygbulup (Misztal et al., 20 14). Alternatively, various models based on marking effect can be used for calculation. The most promising labeling model method is mixed ssHM(Fernando et al., 20 16). If the convergence is proved to be satisfactory, it has enhanced SNP effect and RPG model (Liu et al., 20 14).
When the genome model is based on Gaussian hypothesis, that is, all markers have the same prior variance, there is not much difference between G-based and marker-based one-step models, especially when the residual polygenic effect is assumed and fitted. If different SNP effects can be given different weights in the future, or different SNPs can be fitted for different traits in the multi-trait model, then ssMEM model has obvious advantages over the method of relying on genome relationship matrix G.
One-step evaluation is accused of overestimating the difference in breeding values. Although it is easy to correct the over-prediction of some animals in multi-step evaluation, single-step evaluation is considered as the only evaluation in the population. In order to run the breeding program successfully, the evaluation should rank the animals fairly, whether they are young or old, with or without genomic information and with or without phenotypic information. For the key features in the selected target, over-prediction or over-dispersion is the most serious. Unfortunately, these are also the most urgent features that should be realized in one-step evaluation. This deviation is being deeply studied, hoping to find a general solution to "discount" genome information.
This article comes from a comment: invite comments: unknown-parent groups and meta-founders.
In one-step genome BLUP. (Masuda et al., 202 1).
Introduction:
One-step genome BLUP (ssGBLUP) is a method of genome prediction, which integrates the relationship matrix of family (A) and genome (G) into a unified additive relationship matrix (H matrix), and combines the inverse of this matrix into a set of mixed model equations (MME) to calculate genome prediction.
Question:
The pedigree information of dairy cows is usually incomplete. The loss of pedigree may lead to the deviation and expansion of GEBV obtained by ssGBLUP.
There are three main problems related to pedigree loss in ssGBLUP, namely, biased prediction of selection, inbreeding loss in pedigree relationship and incompatibility between G and A in level and scale.
Solution:
These problems can be solved by using an appropriate model for the unknown population (UPG). The theory behind UPG is true for pure BLUP, but not for ssGBLUP. This study reviews the development of UPG model in BLUP pedigree, the characteristics of UPG model in ssGBLUP, and the influence of UPG on genetic trend and genome prediction.
The similarities and differences between UPG and Metafounder (MF) model (a generalized UPG model) are also commented.
UPG model (QP) transformed from MME has good convergence. However, due to insufficient data, QP model may produce biased genetic trends and may underestimate UPG. The QP model can be changed by deleting the genomic relationship connecting GEBV and UPG effects from the MME. Compared with QP model, this modified QP model has smaller deviation in genetic trend and smaller expansion of genome prediction, especially in the case of large data sets. Recently, a new model was proposed to simulate the pure-bred population, which encapsulated UPG equation into the pedigree relationship of genotyping animals.
MF model is a comprehensive solution to the problem of missing pedigree. If the data set allows to estimate a reasonable MF relation matrix, the model can be used as a choice for multi-variety or mixed evaluation.
Pedigree deletion will affect the genetic trend, but when many confirmed bulls are genotyped, its influence on the predictability of genetic value of genotyped animals should be negligible.
SNP effects can be solved by using GEBV from older genotype animals, and these predicted SNP effects can be used to calculate GEBV of younger genotype animals without parents.