The author is a retired extension dairy scientist in genetics and management and professor emeritus at Virginia Tech, Blacksburg.
Last April, USDA implemented methods to adjust PTAs on genomically tested cows. These PTA adjustments made them more accurate in predicting future progeny performance and more useful in developing future genomic predictions. The changes were applied to all cows with genomic information and those cows with genomic scans imputed from ancestors and progeny. These adjustments created two different genetic scales within female populations because cows with no genomic information or very limited genomic-tested progeny were not adjusted. The result was that PTAs on genomic-tested cows could not be compared accurately to PTAs on cows with no genomic data.
Scientists at the Animal Improvement Programs Laboratory have developed a method to adjust cow PTAs from traditional evaluations to be more like those with genomic predictions. This column discusses how those adjustments are made and how they affect PTAs in Holstein, Jersey, and Brown Swiss cows.
Let's begin by reviewing the problem the adjustments address. It is a significant one for identifying elite females. PTAs on cows and bulls are composed of an estimate of expected merit, measured by two parts:
parent average (PA)
a part due to deviations from that parent average, called Mendelian sampling (MS)
A little more about the science. Think of MS as the difference between a published PTA and the PA on an animal, manipulated in a process called "deregression" to account for differences in the amount of information from progeny and records from cow to cow. Substantial regression of MS causes PTA to be similar to PA for animals with little performance or progeny information. Less regression is used when animals have more records and/or many progeny.
MS on cows is from records and progeny, while MS on bulls is entirely from progeny. Bulls and cows of similar PA should inherit similar genes, on average. Some offspring of similar or identical matings will inherit better gene samples than others. These outliers fuel genetic improvement and we certainly don't want to overadjust their evaluations. The problem is that genetic evaluations for outlier females have been more extreme than for males, even for very similar PAs.
How it works
The new USDA procedure adjusts MS on females to be similar to MS in males. First, PA is subtracted from existing PTAs on cows, producing a "regressed" estimate of MS. Then, reliabilities of PTA and PA are manipulated to deregress MS. Finally, the deregressed estimate is adjusted to have similar standard deviations for both sexes and the adjusted PTA is reconstructed. Let's dig deeper into two key adjustments:
Adjustment varies with reliability of MS. The information available to estimate MS in cows is part of the scaling process. A multiplicative adjustment based on reliability of MS brings standard deviations of MS for cows into line with those for bulls with similar reliability of MS. Cows with extreme estimates of MS based on limited information get the biggest adjustment. Cows with extreme estimates based on lots of daughters, perhaps through ET, are adjusted less because there is more reason to expect genetic differences to cause extreme MS for such cows.
Extreme parent average triggers another adjustment. A second adjustment is based on how different PA on a cow is from the average PA of other animals born in the same year. USDA studies showed that deregressed Mendelian sampling was lower for bulls with superior pedigrees relative to birth year average than it was for cows that were equally superior. The unadjusted PTAs said that cows benefitted more from better parents (within a year of birth) than did the bulls. That's pretty hard to believe.
The adjustment in Holsteins, (-0.434 times difference between a cow's PA and the average PA of all cows born in the same year) is added to the cow's MS. Adjustments are positive for low PAs with negative deviations and negative for high PAs where the deviation was positive and averages zero within a birth year. Larger adjustments, plus or minus, will apply to cows with larger differences between PA and average PA for year of birth.
Adjusted PTAs are allowed to "propagate," meaning that PA of younger animals will be based on adjusted PTAs of dams and other cows in pedigrees. That means adjusted PTAs reflect corrections in ancestor merit, as well as Mendelian sampling. All adjustments are made in a closed, "add-on" system at the end of the regular all-breed animal model programs. Thus, AIPL scientists can modify or eliminate the entire system if better methods are found.
When reviewing the adjustments, cows were grouped into deciles by PA, with over 275,000 cows per group. The largest average change in PTA between adjusted and unadjusted PTAs was in groups 0 and 9.
Bookends affected most
Average change is positive for cows in lower pedigree merit groups and negative for the higher pedigree merit groups. Increases and decreases in PTA reflect the trend line for average adjustment as pedigree merit rises. The largest average drop of 131 pounds of PTA milk was in the top PA group. The range was a drop of 728 pounds and a gain of 524 pounds in the top PA group. Overall, adjustments reduce differences between cows in PTA milk, fat, and protein.
Implementation is scheduled for the April 2011 genetic evaluations. Adjusted PTAs will serve as input for genomic evaluations. Tests show that the adjustments reduce, but do not completely remove, scaling differences between genomically tested cows and genomically tested bulls.
Adjusted PTAs on genomically tested cows will be reduced by another 221, 8.8, and 6.4 pounds for milk, fat, and protein in Holsteins before those PTAs are used in the discovery population for genomic predictions.
Slightly smaller adjustments of 195, 7.9, and 6.9 pounds for milk, fat, and protein will be applied to Jersey/Brown Swiss cow PTAs. Owners of many genomically tested cows have already seen PTAs on their cows adjusted for MS scaling. The new procedures will replace some of that adjustment but are not intended to produce more extreme adjustments.
The bottom line is that PTAs on all cows have been adjusted to be more like evaluations on genomically tested cows. PTAs on cows with genomic evaluations should continue as they have been since April 2010. Generalizations can't cover every situation. I'm sure there will be some cows for which existing PTAs do change for yield traits.
To view the full chart on the size and effects of the adjustments on Holstein cow PTAs using December proofs, check out Dairyman Extras in our Dairy E-Source section.
Last April, USDA implemented methods to adjust PTAs on genomically tested cows. These PTA adjustments made them more accurate in predicting future progeny performance and more useful in developing future genomic predictions. The changes were applied to all cows with genomic information and those cows with genomic scans imputed from ancestors and progeny. These adjustments created two different genetic scales within female populations because cows with no genomic information or very limited genomic-tested progeny were not adjusted. The result was that PTAs on genomic-tested cows could not be compared accurately to PTAs on cows with no genomic data.
Scientists at the Animal Improvement Programs Laboratory have developed a method to adjust cow PTAs from traditional evaluations to be more like those with genomic predictions. This column discusses how those adjustments are made and how they affect PTAs in Holstein, Jersey, and Brown Swiss cows.
Let's begin by reviewing the problem the adjustments address. It is a significant one for identifying elite females. PTAs on cows and bulls are composed of an estimate of expected merit, measured by two parts:
parent average (PA)
a part due to deviations from that parent average, called Mendelian sampling (MS)
A little more about the science. Think of MS as the difference between a published PTA and the PA on an animal, manipulated in a process called "deregression" to account for differences in the amount of information from progeny and records from cow to cow. Substantial regression of MS causes PTA to be similar to PA for animals with little performance or progeny information. Less regression is used when animals have more records and/or many progeny.
MS on cows is from records and progeny, while MS on bulls is entirely from progeny. Bulls and cows of similar PA should inherit similar genes, on average. Some offspring of similar or identical matings will inherit better gene samples than others. These outliers fuel genetic improvement and we certainly don't want to overadjust their evaluations. The problem is that genetic evaluations for outlier females have been more extreme than for males, even for very similar PAs.
How it works
The new USDA procedure adjusts MS on females to be similar to MS in males. First, PA is subtracted from existing PTAs on cows, producing a "regressed" estimate of MS. Then, reliabilities of PTA and PA are manipulated to deregress MS. Finally, the deregressed estimate is adjusted to have similar standard deviations for both sexes and the adjusted PTA is reconstructed. Let's dig deeper into two key adjustments:
Adjustment varies with reliability of MS. The information available to estimate MS in cows is part of the scaling process. A multiplicative adjustment based on reliability of MS brings standard deviations of MS for cows into line with those for bulls with similar reliability of MS. Cows with extreme estimates of MS based on limited information get the biggest adjustment. Cows with extreme estimates based on lots of daughters, perhaps through ET, are adjusted less because there is more reason to expect genetic differences to cause extreme MS for such cows.
Extreme parent average triggers another adjustment. A second adjustment is based on how different PA on a cow is from the average PA of other animals born in the same year. USDA studies showed that deregressed Mendelian sampling was lower for bulls with superior pedigrees relative to birth year average than it was for cows that were equally superior. The unadjusted PTAs said that cows benefitted more from better parents (within a year of birth) than did the bulls. That's pretty hard to believe.
The adjustment in Holsteins, (-0.434 times difference between a cow's PA and the average PA of all cows born in the same year) is added to the cow's MS. Adjustments are positive for low PAs with negative deviations and negative for high PAs where the deviation was positive and averages zero within a birth year. Larger adjustments, plus or minus, will apply to cows with larger differences between PA and average PA for year of birth.
Adjusted PTAs are allowed to "propagate," meaning that PA of younger animals will be based on adjusted PTAs of dams and other cows in pedigrees. That means adjusted PTAs reflect corrections in ancestor merit, as well as Mendelian sampling. All adjustments are made in a closed, "add-on" system at the end of the regular all-breed animal model programs. Thus, AIPL scientists can modify or eliminate the entire system if better methods are found.
When reviewing the adjustments, cows were grouped into deciles by PA, with over 275,000 cows per group. The largest average change in PTA between adjusted and unadjusted PTAs was in groups 0 and 9.
Bookends affected most
Average change is positive for cows in lower pedigree merit groups and negative for the higher pedigree merit groups. Increases and decreases in PTA reflect the trend line for average adjustment as pedigree merit rises. The largest average drop of 131 pounds of PTA milk was in the top PA group. The range was a drop of 728 pounds and a gain of 524 pounds in the top PA group. Overall, adjustments reduce differences between cows in PTA milk, fat, and protein.
Implementation is scheduled for the April 2011 genetic evaluations. Adjusted PTAs will serve as input for genomic evaluations. Tests show that the adjustments reduce, but do not completely remove, scaling differences between genomically tested cows and genomically tested bulls.
Adjusted PTAs on genomically tested cows will be reduced by another 221, 8.8, and 6.4 pounds for milk, fat, and protein in Holsteins before those PTAs are used in the discovery population for genomic predictions.
Slightly smaller adjustments of 195, 7.9, and 6.9 pounds for milk, fat, and protein will be applied to Jersey/Brown Swiss cow PTAs. Owners of many genomically tested cows have already seen PTAs on their cows adjusted for MS scaling. The new procedures will replace some of that adjustment but are not intended to produce more extreme adjustments.
The bottom line is that PTAs on all cows have been adjusted to be more like evaluations on genomically tested cows. PTAs on cows with genomic evaluations should continue as they have been since April 2010. Generalizations can't cover every situation. I'm sure there will be some cows for which existing PTAs do change for yield traits.
To view the full chart on the size and effects of the adjustments on Holstein cow PTAs using December proofs, check out Dairyman Extras in our Dairy E-Source section.
