A year ago we introduced this topic with what was known at the time. Since then, we are seeing varying levels of AI stud adoption, from using G tests only to choose young sires, to assimilating G tested sires right into their main sire lineups as if they were fully “proven”. We have some observations drawn from recent conversations around the world.
So how have G tested sires fared in New Zealand (where G tested sires hit the market in 2005) ?
The two major AI systems in New Zealand are LIC New Zealand (who breeds 80% of the cows) and CRV Ambreed (who breeds 10% of the cows). Looking at the top six rated sires for each AI system over five years of progeny produced (from 2001 through 2006 calvings), the following happened:
Overestimation on EBV Protein: Holstein: 25% for LIC, 11% for CRV. Jersey: 35% for LIC, 19% for CRV.
Overestimation on BW index: Holstein: 30% for LIC, 11% for CRV. Jersey: 20% for LIC, 9% for CRV.
These twelve bulls (each breed) sired a total of 500,000 progeny annually that were tested. So Rel% for all sires (assigning 75% to the G tests) ended up at 99% Rel (progeny data replacing the G estimates).
These results (consistently overstating the superiority of the sires G tested as at “the top”) have driven the Kiwi’s back to their microscopes and computers to figure out why Genomics failed as a predictor. One of their conclusions was that higher density SNPs were required… and always, more data…
So what are the latest developments incorporated into the August 2009 predictions here?
An individual cow of interest (who has produced a polled Red & White son) has the following history of Genomic evaluations: April 2009: GPTA $251 Net Merit: August 2009: GPTA $453 Net merit!
Over $200 gained in “Net Merit” rank, without any change in her type scores or lactation records.
What changed? After all, her Genomic SNP (direct measurement of gene enzymes and sequences) would be the same (genes never change from conception to demise). But this particular cow has a German AI sire—foreign G test results were not included by USDA in April, but were in August. In other words, although the G test is a “direct” look at the individual genotype, we still are interpreting the results in part, on the parentage of the animals tested.
Should not the accumulating body of G tests, establishing “marker” genes for each evaluated trait, be the determinant of a Genomic value, rather than a continuing reliance on the simplistic (lower Rel%) Parent Average calculations? After all, we had already determined “G” tests were 70% Rel on milk and 60% Rel on type, on the first go-rounds (January 2009).
I posed this question to Dr Curt Van Tassel, of AIPL-USDA, who is deeply involved in the computer estimation of trait values from Genomic testing. His point gave clarity to why we could see such a change on a cow: “…we can only predict genetic values accurately for animals that are represented in the data. What I mean… is that if we have not seen an animal with a genotype like the one we are trying to predict, we don’t have a great ability to predict that genetic value.” The less common a pedigree, the more estimation falls back on Parent Averages.
This goes a long way toward explaining why we keep seeing “the same” pedigree combinations on those G tested young sires who have the “elite” PTA estimations. Those pesky genes still do not wear name tags that say, “I am the milk gene…” Thus USDA researchers are hoping to implement new genotype platforms with higher density (600,000 SNPs compared to the current 50,000 SNPs).
Dr Van Tassell’s “quotes” come from the online “polled dairy cattle” discussion group within his responses to questions.
What conclusions did Dr Van Tassel offer for the market use of Genomic data ?
“I was asked after a presentation that I gave to the New York all-breed society meeting in January 2009 whether I would use bulls based [entirely] on Genomic testing. My response (which I stand by here in August) was that if I had spent my life building a herd of cows, that I would look at any technology that had the ability to undermine those efforts with great skepticism. I think that this technology has great promise, but as of yet, is still largely untested in the true application of predicting response to selection using genomic predictions.”
He earlier answered my query, “that I am absolutely correct that we have not seen any real evidence of the PREDICTIVE power of the Genomic PTA. Everything up to now has been predicting genetic merit for animals that were selected using quantitative genetic tools that we then re-predicted using SNP data, OR animals that have had genomic predictions early in life that have not been validated by real progeny data.”
The current excitement over Genomics by those scientists involved at AIPL is… “we have recreated [the selection scenario] by using genomic data from historic bulls to predict future genetic merit using data of 5+ years ago, and then looked to see if the predictions were more accurate than the old [pedigree-based] evaluations, and they indeed were!”
What underlying problems plague accuracy in genetic evaluation?
Dr Van Tassel reminded us that based upon DNA samples of heifers given ID, anywhere from 10% to possibly 40% of the progeny lists of sampled sires are misidentified (by sire or dam or both). Because this has an ongoing potential to cause fluctuation in progeny evaluations, it costs us all in accuracy of the data we attempt to use for genetic selection improvement.
How should we then adapt G tested sires in our breeding programs?
They remain, for practical purposes, “young sires for sampling”. Using a group rather than a single bull choice; using added data from their individuality, pedigree and aAa; avoiding paying too great a premium for an elite G-based ranking level; all this makes common sense.
Based on this advice from a scientist actively working in Genomics (rather than a magazine writer who needed an upbeat topic for the latest issue) – we offer G tested sires as “super samplers”, and we do not confuse the ranking of progeny-tested sires on our price lists alongside G tested young sires.
A final quote from Dr Curt: “if this technology does work, then there is a huge opportunity for someone to use it aggressively as an ‘early adopter’.” So do your homework and buy sires accordingly.
The breeder’s philosophy in the use of sire selection tools
We tend to believe that, every time a new technology comes along, it makes earlier practices obsolete. In dairy cattle selection and mating, this is not supported by experience. The more tools available, old or new, the better job we do at sorting the long-term useful (dependable) from the short-term novelty.
Genomics is no different today than indexes were in 1970; they are a way to screen animals, while you seek the traits and qualities specific to needs you have in your herd. The identification of those traits specific to problems in your herd, which can be solved by careful analysis and mating by heritability, remains more important to your profitability than any external ranking of genetic value.
Are you ready to pass up progeny testing to rely completely on Genomic testing?
At least one major AI system is so convinced that Genomics is a “done deal” that it has replaced 99% Rel progeny-tested sires with G-tested sires on its active sire price list and is scaling back its young sire sampling program in favor of its own ET donor herd.
Inside you will find data of earlier G-test experiences around the world, an example of a current issue in debate on how G testing should be done, and a recommendation from a key researcher that cautions caution alongside optimism.
Our goal is that your herd continue to improve as fast as good genes and sound matings and competent heifer rearing allows. We hope you find this information challenging and able to stimulate better-informed questions on sire selection.