Abstract :
[en] Along with technical developments, international exchanges of genetic material
(e.g., frozen semen, embryos) have increased since the 1970s. However, genetic
evaluations are traditionally based on phenotypic and genealogical data which are
internally recorded, i.e., within well defined borders. Because imported (i.e., external)
genetic material is usually strongly selected in their respective populations, internal
genetic evaluations for external animals could be biased and less accurate if external data
used for their selection is ignored. Moreover, comparison of internal and external animals
based on their internal and external estimates of genetic merit is needed to select and
potentially import the most suitable ones according to the internal breeding goal.
However, such comparison is usually not possible among internal and external genetic
evaluations due, e.g., to differences among units of measurement. Thereby, several
approaches and algorithms have been developed to render internal and external genetic
evaluations comparable, and to combine or blend phenotypic and genealogical data and
external information, i.e., estimates of genetic merit and associated reliabilities.
Furthermore, the recent development of genomic selection also increased needs for
combining phenotypic, genealogical and genomic data and information. Therefore, the
aim of this thesis was first to develop innovative algorithms to combine diverse sources of
phenotypic, genealogical and genomic data and information, and second to test them on
simulated and real data in order to check their correctness. Based on a Bayesian view of
the linear mixed models and addressing several issues highlighted by previous studies,
systems of equations combining simultaneously diverse sources of data and external
information were developed for (multivariate) genetic and single-step genomic
evaluations. Double counting of contributions due to relationships and due to records
were considered as well as computational burden. The performances of the developed
systems of equations were evaluated using simulated datasets and real datasets originating
from genetic (genomic) evaluations for Holstein cattle and for show jumping horses. The
different results showed that the developed equations integrated and blended several
sources of information in a proper way into a genetic or a single-step genomic evaluation.It was also observed that double counting of contributions due to relationships and due to
records was (almost) avoided. Furthermore, more reliable estimates of genetic merit were
also obtained for external animals and for their relatives after integration of external
information. Also, the developed equations can be easily adapted to complex models,
such as multivariate mixed models. Indeed, it was shown that external information
correlated to the internal phenotypic traits was properly integrated using the developed
equations. Finally, research of this thesis led to the development of a genomic evaluation
system for Holstein cattle in the Walloon Region of Belgium for production traits, as well
as for other traits, like somatic cell score. Based on the research of this thesis, future
research topics, e.g., concerning integration of correlated external information and of
genomic information, were finally presented.