The accuracy of Genomic Predictions (GP) in small populations especially in developing countries and breeding companies is constrained by; terms of the structure of the reference and validation populations, response variables, genomic prediction models, validation methods.
For numerically small breeds, setting up a large reference population for GP is hard and thus to overcome the lack of reference data the following strategies can be employed (Schöpke and Swalve (2016); Gholami et al. (2021); Mrode et al. (2019));
- The use of combined reference populations from different breeds, different countries, or different research populations.
- Including female information into the reference population.
- Imputation of un-genotyped animals.
- Collaboration between developing (with small reference populations) and developed countries is important in implementing genomic breeding technologies in the former, especial in dairy cattle, where there has been a large importation of bulls. It is likely that most of these bulls have been genotyped in the developed countries and willingness to share genotypes and some other relevant performance data will help in enlarging the reference population and hence the accuracy of genomic predictions in developing countries.
Additionally, the use of a multi-breed GRM model (MBMG) makes it possible to use information from numerically large breeds to improve prediction accuracy of numerically small breeds. The superiority of MBMG is mainly due to its ability to use information on pre-selected markers, to explain the remaining genetic variance and weigh information from a different breed by the genetic correlation between the breeds (Raymond et al. (2018)).
Currently, the statistical approaches that provide the highest accuracy for genomic evaluation of small populations involve the use of two types of models and they include:
1. The single-step genomic best linear unbiased prediction models
2. The single-step Bayesian models
The validation of genomic evaluations using the single-step methods is done using the LR method (Legarra and Reverter (2018)).
References