Dairy Evaluation Changes

In the December evaluation, ICBF implemented new software to calculate genomic predictions and reliabilities.  The new software ensures the genomic predictions are utilising the most up-to-date methods available which replaces the old system which had been in operation since 2009.

The new software now evaluates 54,000 SNP (Single nucleotide Polymorphism) effects compared to 40,000 SNP effects in the old system. The additional 14,000 SNP effects allow for a 35% increase in coverage of the genome for the new evaluation.

Updated training population

Implementation of the new software has allowed ICBF to   increase the training population used for the predictions by including additional, informative animals. These include AI sires and natural service sires which have daughter proven proofs. It was previously not possible to include these animals as the old software could only incorporate animals with a reliability over 70%. An additional 2,216 AI and natural service sires have been added to the training population for milk traits (milk kg, fat and protein), and an additional 1,419 sires have been added for fertility, giving a total population of 7,473 animals in the training dataset for production traits and 1,419 for fertility traits.

New traits

Genomic PTA values are now being produced for the first time for 4 management and health traits, namely mastitis, lameness, milking temperament and milking speed.

Reliability

The new software calculates reliability in a different way to the old. The old software was considered best practise in 2009 (when genomic evaluations were launched), but newer, more sophisticated methods have been developed since then.

There are numerous accepted ways of calculating reliability. However, all implemented methods at evaluation centres worldwide are an approximation, as computation of exact reliabilities is a very computationally demanding exercise.

The new software has identified two areas in the calculation of reliabilities which needed to be recalibrated.

  • Avoidance of double counting.

It is now known to be important for training sires to also account for the number of sons of these sires that are also in the training population. These sons are expressing the DNA signature of the sire already and thus correction needs to be made for this to avoid double counting.

  • Number of SNP effects analysed.

It is now known that estimation of reliability is dependent on the number of SNP effects analysed. Inclusion of additional SNP effects will establish more relationships between animals. Establishing more relationships will result in a reduction in reliability for some young sires, who now appear to be more closely related to the reference population. This is closely linked to the double counting effect in point 1.

 Future Benefits

The new software will also means that genomic proofs for breeds such as Jersey, Norwegian Red and crossbreds, are a more realistic target in the near future.