Genomic Prediction
Genomic prediction lies at the heart of modern breeding.
Breeding organisations rely on prediction models to estimate breeding values and guide selection decisions on a daily basis.
In Work Package 4, we focus on developing reliable, scalable and well validated prediction methods that can support these decisions in practice. This includes ensuring that models remain robust as data volumes grow and new types of information become available.
Why it matters
Reliable genomic prediction is essential for effective breeding decisions.
As breeding programs generate larger and more complex datasets, the challenge is no longer only to build prediction models, but to ensure that they remain reliable, interpretable and applicable in practice.
Data limitations, structural issues in datasets and increasing model complexity can all affect the quality of predictions.
If these challenges are not properly addressed, there is a risk that prediction results are less robust, making it more difficult to rely on them in routine decision making.
What we do
Within Work Package 4, we develop and deliver state of the art methods and software for genomic prediction in modern breeding programs.
At the core of this work is the continuous development and optimisation of the MiXBLUP software suite. These tools are used by breeding organisations for large scale genetic evaluations and must therefore be fast, reliable and robust in daily practice.
We focus on improving not only the performance of prediction models, but also their stability and usability. This includes enhancing error handling, supporting implementation in routine evaluations, and ensuring that tools can be applied consistently across different datasets and populations.
A key part of this work is the development of validation and diagnostic tools. These allow users to better understand the quality and limitations of their data, for example by identifying issues such as confounding or insufficient connectedness. This ensures that predictions are not only accurate, but also trustworthy.
In addition, Work Package 4 develops new modelling approaches that integrate intermediate features and additional sources of information. By combining phenotypic data with emerging data types, we aim to further improve prediction accuracy and better reflect biological complexity.
Work Package 4 works closely with the other Work Packages, in particular to incorporate structural variation into prediction models and to support decision making in breeding program design.
Impact
Work Package 4 enables breeding organisations to rely on robust, scalable and well validated genomic prediction tools.
By improving both the accuracy and the reliability of predictions, it becomes possible to make better informed selection decisions in routine breeding practice. At the same time, improved diagnostics and validation ensure that results can be interpreted with confidence.
The continuous development of software and methods allows organisations to keep pace with increasing data volumes and new types of information. This supports more efficient use of data and ensures that genomic prediction remains a strong foundation for modern breeding programs.
Ultimately, Work Package 4 translates complex data into reliable breeding values, enabling consistent and effective decision making at scale.