Multi-breed Genetic Evaluation (6)
International Genetic Solutions (IGS) is an unprecedented collaboration between progressive breed associations fervently committed to enhancing commercial profitability. The collaboration has yielded the world’s largest genetic evaluation of beef cattle with over 17 million animals and 120,000+ genotypes.
In keeping with our commitment to the cattle industry, IGS is pleased to announce the IGS Multi-breed Genetic Evaluation powered by BOLT. The new genetic evaluation provides more predictive EPDs, better use of genomics, more accurate accuracy reported with EPDs, all with weekly evaluations. The announcement ushers in a new era in genetic evaluation — an era made possible by a genetic evaluation system dubbed BOLT (Biometric Open Language Tools, owned by Theta Solutions, LLC).
The concept for BOLT started in 2014 as a research endeavor between the American Simmental Association and Drs. Bruce Golden and Dorian Garrick. BOLT is, quite simply, the most revolutionary and powerful genetic evaluation system in existence. Its power allows IGS to leverage genetic evaluation methodology that was once thought to be untenable on large databases — methodology that significantly improves genetic prediction.
In December 2016, IGS published a multi-breed stayability, the industry’s first EPD using BOLT and the first single-step methodology applied to a large beef cattle database. Since that time, the IGS genetic evaluation team has worked toward fully implementing BOLT with an automated system that enables weekly evaluations for an entire suite of EPDs. As of May 5th, 2018, ASA is the first of the IGS partners to publish a full suite of EPDs generated by the IGS Multi-breed Genetic Evaluation powered by BOLT. Each IGS partner has complete autonomy to determine the release date that best fits their organization. As such, the release of EPDs by the other IGS partners is likely to be staggered over the next several weeks. As always, we look forward to your questions and comments about what you see.
Here are the notable changes in the evaluation:
Movement of EPDs and reranking. EPDs and indexes will change. These changes will be more dramatic for younger, lower accuracy cattle. The IGS team has tested the changes and proven the new EPDs result in superior predictions of genetic merit.
Shrinking of EPD range. You will notice a reduction in the range of EPDs for most traits. The IGS evaluation team tested the statistical veracity of the reduction and it has proven to be in line with expectations based on the genetic variation in the population.
Improved use of genomics. With the switch to the BOLT software, IGS will use single-step genomic evaluation on all EPDs. Single-step uses DNA markers, pedigree information, and phenotypic data simultaneously in the prediction of EPDs. Previously, molecular breeding values (MBVs) were calculated from the genomic information and those MBVs were blended in a separate procedure into the EPD predictions. The single-step method squeezes more information from the DNA markers than the previous approach allowed. Additionally, with single-step, the genomic information will not only enhance each EPD for the genotyped animals but also will be used in the EPD estimates of relatives.
The table below shows how many progeny records it takes for an animal without genomics to have the same BIF accuracy as the young animal with genomics (but no progeny). In other words, EPD on a genotyped 1-month-old calf will be as accurate as an animal with birth weights on 21 calves, weaning weights on 22 calves, etc. The carcass traits represent actual carcass records, not ultrasound records. You may notice the maternal calving ease gets the least boost from genomics. This is due in part to such few females being genotyped.
It is important to note, continued collection of phenotypic records remains a vital part of genetic predictions. DNA testing will never replace the need to record and submit phenotypes.
It is well established that DNA markers vary greatly in their effect on traits — ranging from large to virtually no impact. To leverage this biological fact in a statistically advantageous manner, the BOLT single-step method only uses markers that have a meaningful impact on the traits of interest, while ignoring those that have little to no effect. Research has shown that by using this approach, BOLT reduces statistical “noise” and thereby increases the accuracy of the EPD prediction compared to other single-step methods.
More accurate accuracy. In the previous IGS evaluation platform and all others in existence other than BOLT, the calculation of the accuracy associated with each EPD is achieved through “approximation” methods. It has long been known these methods are a less than optimal approach to the calculation of accuracy — tending to overestimate accuracy. By employing unique computing strategies that leverage both software and hardware efficiencies, BOLT performs what was previously unthinkable — utilizing a sampling methodology to calculate what is essentially true accuracy. Unlike approximated accuracies, BOLT-derived accuracies will result in predicted movements associated with possible change holding true over time. This is not the case with the previous IGS software or any other system currently in existence.
While the IGS evaluation team and partners are excited to release this new chapter in genetic evaluation, the new genetic evaluation system will only realize its true potential if the selection is made using its EPD and index values. Hands down, there is no better (more accurate) way to select for quantitative traits than an EPD. Economic indexes predict net profit by weighing the EPD for economically relevant traits coupled with economic estimates. To compete with other protein sources, it is imperative that the beef industry adopts the best science and technology to make better breeding selection decisions.
Please note, each IGS breed association has the latitude to publish the BOLT generated EPDs when the timing is right for their association.
For more information about the IGS Multi-breed Genetic Evaluation powered by BOLT, go to www.internationalgeneticsolutions.com.
The new genetic evaluation, Multi-breed Genetic Evaluation powered by BOLT, offers groundbreaking advances in the prediction of EPDs for the IGS group. Here are some frequently asked questions and answers to help you better understand Multi-breed Single-step.
1. What are the key features of the Multi-breed Genetic Evaluation powered by BOLT?
• Faster and more automated system allowing for frequent genetic evaluations.
• Improved use of genomic data.
• Improved methodology for predictions of all traits.
• More accurate accuracy.
• More flexibility to add additional traits or change methods for future improvements.
2. How is ASA’s single-step approach different from blending for genomic evaluation?
The blending approach uses separate steps to calculate genomically enhanced EPDs. This approach requires two steps. The first step is to estimate the effects of DNA markers through a process called “training” or “calibration”. These effects are then used to calculate molecular breeding values (MBVs) on genotyped animals. The MBVs are then combined with traditionally calculated EPDs to enhance the accuracy of the traditionally calculated EPDs. The blending process is only performed on genotyped animals.
Befitting its name, the single-step approach calculates genomically enhanced EPDs in one step — using DNA, pedigree information, and phenotypes simultaneously. As a result, the DNA information not only improves the accuracy of prediction on genotyped animals, but also on the relatives and contemporaries of the genotyped animals. In a sense, all animals are genomically enhanced under the single-step approach.
There are also issues inherent in the blending process that are solved with single-step. Similar to the fact that only reporting phenotypes on a selected group of animals in your herd can lead to less informative (and more biased) EPDs with traditional evaluation, problems can exist with blending as it only involves genotyped animals — and genotyped animals tend to be highly selected. However, because single-step includes information from non-genotyped as well as genotyped animals, the issues are corrected.
3. How is the Multi-breed Genetic Evaluation powered by BOLT different than other single-step models used in other genetic evaluations?
It is well established that DNA markers vary greatly in their effect on traits — ranging from a large to no impact. To leverage this biological fact in a statistically advantageous manner, the BOLT single-step method only utilizes markers that have a meaningful impact on the traits of interest, while ignoring those that have little to no effect. By using this approach, BOLT reduces the statistical “noise” and thereby increases the accuracy of prediction. By circumventing the “noise,” BOLT-generated EPDs tend to be more accurate than EPDs generated by organizations that are relegated to using all markers in their single-step evaluation.
4. How many DNA markers are being used?
The Multi-breed Genetic Evaluation powered by BOLT uses a subset of weighted markers based on a research study performed by Drs. Mahdi Saatchi and Dorian Garrick, while they were scientists at Iowa State University. Drs. Saatchi and Garrick first used the 50,000 markers to determine a subset of weighted markers that are highly associated with economically relevant traits in beef cattle with consistent effects across breeds. Because the IGS evaluation is for multiple breeds, it is important to remove markers with inconsistent effects or no effects in different breeds.
The Saatchi and Garrick research also found that utilizing genotypes on animals of multiple breeds consistently increased the accuracy of prediction within a particular breed when compared to limiting DNA utilization to only animals of a particular breed.
5. Why are some traits influenced by markers and others are not?
The genetic architectures of various traits are different. Some are controlled by few genes with large effects and some are controlled by many small effects genes. In the current DNA profilers, there are some markers with high correlations with corresponding genes for some traits and low correlations with others. That’s why we see the different DNA added values for different traits. It is hard to change the genetic architecture of a trait. But, new DNA profilers or future technologies may help to improve the value of DNA information for such traits. Furthermore, some maternal traits, like Maternal Calving Ease and Milk, are difficult to predict with genomics because there are so few females genotyped. Increasing the number of cows and heifers genotyped will improve the ability to use genomics to predict maternal traits.
6. Will genomic testing replace the need to submit phenotype records?
No, reporting actual records is critical. The value of genomic predictions increases as the amount of phenotypic information increases. Furthermore, at this point, animals cannot achieve high accuracy with genomic data alone. High accuracy EPDs are only achievable by collecting many phenotypic records on offspring.
7. How do we know predictions via BOLT are better than the previous system (Cornell software)?
The IGS evaluation team has conducted a series of validations to compare the BOLT system to the Cornell system. BOLT-derived EPDs had higher correlations to birth, weaning and yearling weights (0.34, 0.29, and 0.26, respectively) than the Cornell derived EPDs (0.27, 0.19, and 0.20, respectively). Furthermore, there was a larger difference in average progeny performance (birth, weaning, and yearling) of the top 1% compared to the bottom 1% animals in the BOLT derived EPDs compared to the Cornell calculated EPDs. Both validations suggest the BOLT EPDs align better with the actual phenotypes than the Cornell EPDs.
8. Why do some animals have substantial changes in their indexes?
Though the correlations between the previous (Cornell derived) EPDs/indexes and the BOLT derived EPDs/indexes are relatively strong, there will be some animals that happen to move in a consistently favorable or unfavorable direction in a number of EPDs. Because indexes are comprised of several EPDs, even though movement in individual EPDs may be considered small, movement in the same direction across EPDs may yield sizable movements in the index value. This is particularly true for animals that have consistent movement in traits that are drivers of a particular index. Though in a large population like ours we would expect to see several animals with substantial index movement, these animals will be the exception to the rule.
9. How does BOLT improve our calculation of accuracy?
“True” accuracy can be thought of as the gold standard of accuracy. It is statistically unbiased, and therefore the ultimate measure of accuracy. True accuracy is the accuracy resulting from direct calculation. Unfortunately, even with the massively powerful computing capacity now in existence, the direct calculation of accuracy is not possible on datasets the size of ours. Because we cannot calculate accuracy directly, other approaches to accuracy calculation have been developed.
In our Cornell evaluation platform and all others in existence other than BOLT, the calculation of the accuracy associated with each EPD is achieved through “approximation” methods. It has long been known these methods are a very crude approach to the calculation of accuracy — tending to overestimate accuracy.
Another approach to the calculation of accuracy is via “sampling” methodology. Sampling is shown to be a more accurate predictor of accuracy. In fact, the results of this method were reported to be virtually identical to true accuracy. Unfortunately, due to its computationally intense nature, sampling has long been thought an infeasible approach to the calculation of accuracy on large databases.
BOLT, however, has changed the landscape in this area. By employing unique computing strategies that leverage both software and hardware efficiencies, BOLT performs what was previously unthinkable — utilizing a sampling methodology to calculate what is essentially true accuracy.
Because BOLT can calculate true accuracy, we can put more confidence in our accuracy metrics. Put another way, unlike with approximation, we can count on the predicted movements associated with possible change holding true over time. This was not the case with our Cornell system nor any other system in existence.
10. Why do the carcass EPDs generally have an increase in accuracy with BOLT while this is not a case for other traits?
You will notice that while the Multi-breed Genetic Evaluation powered by BOLT will generally produce lower accuracies than the Cornell system for growth and calving ease traits, the opposite is true for carcass traits.
One reason behind the differing accuracy outcomes is several years ago ASA staff developed a way to temper inflated accuracies in the Cornell carcass evaluation. Unfortunately, this was not possible for growth traits.
Another reason is that the Cornell system only used the carcass and its corresponding ultrasound trait (e.g., marbling score and IMF) to predict carcass EPDs, while records on several additional correlated traits are leveraged with the BOLT system.
A new feature of the BOLT evaluation is a new approach to the calculation of Carcass Weight EPDs. Due to limitations, our previous Carcass Weight EPDs did not incorporate actual carcass weights. They were predicted through an index of birth, weaning, and yearling weights. Besides using prior growth records (weaning, post weaning), the new approach also includes actual carcass weights. This feature will undoubtedly lead to a more accurate prediction of carcass weight.
11. What can I do to improve the predictions on my herd?
Whole Herd Reporting — If you haven’t already, you should consider enrolling your entire herd with a breed association total herd reporting program as it offers the most complete picture of the genetics involved in your herd.
Proper contemporary groups — It is important for the genetic evaluation that you group, to the best of your ability, animals that were treated uniformly. Proper reporting of contemporary groups ensures better predictions for all.
Take data collection and reporting seriously — Phenotypes are the fuel that drives the genetic evaluation. Take pride in collecting accurate data. If possible, try to collect additional phenotypes like mature cow weight, cow body condition score, feed intake, and carcass data.
Use genomics — DNA testing adds more information to what we know about an animal. The more genotypes we collect, the better we can predict DNA-tested animals in the future. Also, the more relatives genotyped, the better we can predict their relatives in future generations. Therefore, to ensure your bloodlines are well represented in the predictions, genotype your animals.
PDF versionPDF version
DNA profiles provide additional information about the genetic merit of a DNA tested animal and increase the accuracy of EPDs, which are called Genomic Enhanced EPD or GE-EPDs. In the IGS Single-step process, the DNA marker genotypes are directly incorporated into the genetic evaluation along with the phenotypes (performance data) and the pedigree. As a result, the DNA information has an impact not only on the genotyped individual but also on all the relatives of that genotyped individual. This allows for the DNA information to improve the accuracy of non-genotyped relatives. To measure the impact of DNA information on accuracies of GE-EPDs in the IGS Singlestep genetic evaluations, we compared the average BIF accuracies of GE-EPDs of DNA tested young animals (born in 2016 with no progeny) to the average BIF accuracies of nongenotyped sires born in 2010-2014. Only sires with non-genotyped calves were used for this comparison. We found that the average BIF accuracy of GE-EPD for a DNA tested young animal is equivalent to the average BIF accuracy of a non-genotyped sire with 21, 22 and 24 calves with observed phenotypes for birth, weaning and yearling weights, respectively (Figure 1, where a horizontal line cross a curve for a specific trait (e.g. red line and blue curve cross each other at the data point correspond to y(accuracy)=0.46) and x(progeny)=21 for birth weight)). The progeny equivalent (PE) for direct calving ease was 15 and it was only 3 for total maternal calving ease due to limited genotypes on cows. The PE for milk and stayability were 18 and 25, respectively (Figure 1).
The Multi-breed Evaluation powered by BOLT is a breakthrough in GE-EPD accuracy improvement. Enabling technologies such as BOLT software will allow for even faster genetic progress with more accurate EPDs earlier in an animal’s life. We (IGS) are dedicated to using the best available technology to deliver more accurate GE-EPDs to our members so they have the best tools available for their selection decisions.
By Mahdi Saatchi, Rohan L. Fernando, Lauren Hyde, Jackie Atkins, Steve McGuire, Wade Shafer, Matt L. Spangler, and Bruce Golden, IGS Genetic Evaluation Team, and Consultants.
The ASA and International Genetic Solution (IGS) partners invested in a new and improved genetic evaluation software called BOLT to replace the Cornell EPD evaluation system. Among other benefits, this enables the use of Single-step methods for incorporating genomic information into the National Cattle Evaluation instead of the blending approach. In the Single-step process, the DNA marker genotypes are directly incorporated into the genetic evaluation along with the phenotypes (performance data) and the pedigree. As a result, the genomic data has an impact not only on the genotyped individual but also on all the relatives of that genotyped individual. This allows for the genomic information to improve the accuracy of non-genotyped relatives.
The IGS Multi-breed Single-step powered by BOLT squeezes more information from the DNA markers by allowing for certain DNA markers to have a larger influence on predicting the genetic merit of an animal than other DNA markers while some DNA markers to have no effects on trait(s) of interest (for progeny equivalents of select traits, see page 46). This model is closer to what we expect based on biology where some parts of an animal’s genome (or genes) play more important roles than other parts of its genome (or genes). This is unique to the IGS Single-step method compared to other organizations where the DNA marker information is used to adjust relationships among the individuals.
Many ASA members and IGS partners wonder if the BOLT EPDs are more accurate than the Cornell derived EPDs in the real world? To answer this question, we performed a validation study where we ran a data set (pedigree, performance, genomics) through both genetic evaluation software (BOLT and Cornell) to compare the accuracies of the EPDs produced. To enable a fair comparison, we removed the performance records of animals born in 2015 and later from the evaluation in both systems to be used as progeny performance records for validation purposes. Table 1 shows the correlations between EPDs and progeny performance of non-genotyped sires evaluated in both systems that have progeny born in 2015 or later with recorded birth, weaning, and yearling weights. As shown, the BOLT EPDs are more accurate than Cornell EPDs as the correlations are higher for BOLT EPDs with sires’ progeny performances.
Table 1- The correlations between BOLT vs. Cornell EPDs with progeny performance of non-genotyped sires for birth, weaning and yearling weights.
|Trait 1||N of Sires||BOLT||Cornell|
To have a better sense of improvement in accuracies, we ranked sires based on either BOLT or Cornell EPDs for birth, weaning and yearling weights. Then, we compared the progeny performance of the top 1% vs bottom 1% ranked sires for each trait in each evaluation system. The results are shown in Table 2.
Table 2 – The average progeny performance of non-genotyped sires ranked based on either BOLT or Cornell EPDs
|Cornell||BOLT vs Cornell|
|Trait||N of sires||Top 1%||
|Difference||Top 1%||Bottom 1%||Difference||Top 1%|
As you can see, the BOLT EPDs ranked sires more accurately than EPDs from the Cornell software, where progenies of top 1% ranked sires based on the BOLT EPDs are +3.1, +16.8 and +40.2 lb heavier at birth, weaning and yearling. These results are exciting and show that our investment in new technology will lead to more accurate EPDs.
Change can be a scary concept to some yet sought after by others. Many ASA members and International Genetic Solution (IGS) partners wonder about the changes on the horizon once the new evaluation, Multi-breed Evaluation powered by BOLT, is fully implemented. That change may be nerve-racking but in reality, things should change. Why invest in new and improved methods if you get the same answers? Here are key changes to expect with the new genetic evaluation:
1. Movement of EPDs and reranking. EPDs will change especially in younger, lower accuracy cattle. Members should expect movement in lower accuracy cattle, as seen in the existing evaluations, because they may have new progeny data reported. Some cattle will move in a favorable direction while others will do the opposite. Keep in mind even if the EPDs get worse, the prediction of them is more accurate. With Multi-breed Evaluation, we will have more accurate EPDs earlier in an animal’s life.
2. More accurate accuracy. This idea takes a little time to sink in. The accuracy reported for each EPD will be more directly calculated and thus closer to the “real” accuracy. The methods to solve accuracy directly are extremely difficult and take a lot of computer power. With the previous Cornell software, it was not possible to solve for accuracy directly so an approximation method was used to estimate accuracy for each EPD. There were inherent flaws with approximating the accuracy of the previous method. Now with BOLT software, the accuracy reported with the EPD will be more reliable.
3. Reported accuracies will tend to be lower. One of the inherent flaws in the approximation methods used to find accuracy in the previous evaluation, and in all evaluations not produced through BOLT, was they frequently overestimated accuracy, especially for younger animals. This was known for a long time, but there was no way to calculate the accuracies directly. With BOLT, having accuracy more directly solved results in a more reliable accuracy but that accuracy will often be numerically lower than the previous evaluation would predict. However, the newly reported accuracies with BOLT should better represent the possible changes for the EPDs.
4. DNA testing will have a larger impact. With the switch to BOLT software, IGS will use Single-step genomic evaluation on all EPDs (currently using Single Step for Stayability EPDs). Single-step uses the DNA markers, pedigree information, and phenotypic data simultaneously in the prediction of the EPDs. Previously molecular breeding values (MBVs) were calculated from the genomic information and those MBVs were blended separately into the EPD prediction. The Single-step method squeezes more information from the DNA markers than the previous approach allowed. Also, there are biases inherent in the blending process that aren’t a problem with the Single-step approach. Additionally, with Single-step, the genomic information will not only enhance EPDs for the genotyped animal but also will be used in the EPD estimates of relatives.
5. Weekly genetic evaluation runs. With the horsepower behind BOLT, IGS can run genetic evaluations weekly. This has many benefits. It allows members to get more immediate feedback after submitting their records. If members miss a deadline, they can catch the next evaluation run the following week. It allows for more accurate EPDs throughout the year and faster incorporation of the genomics. This also means the EPDs put in print will quickly be outdated.
Genetic evaluation is not stagnant. There will always be improvements as new research in animal breeding, genomics, and statistics advance. BOLT software is revolutionary in the innate flexibility, the computational power, and the statistical methods made possible using this software. IGS Multi-breed Single-step powered by BOLT promises more accurate EPDs, accuracies, and better use of genomics all delivered to you on a weekly basis.
By Drs. Jackie Atkins, Lauren Hyde, and Wade Shafer
It’s here! The Spring 2017 evaluation includes a multi-breed Stay EPD using the much anticipated new genetic evaluation software, dubbed BOLT (Biometric Open Language Tools). The Stay EPD uses a random regression approach to predict stayability developed by Dr. Janusz Jamrozik and colleagues from the University of Guelph and the Canadian Simmental Association. Dr. Scott Speidel at Colorado State University, Dr. Bruce Golden (Theta Solutions), and the genetics team at International Genetic Solutions (IGS) (Lauren Hyde, Wade Shafer, Mahdi Saatchi, and Steve McGuire) worked together to incorporate the new methodology into the IGS evaluation. The new stayability evaluation includes data on animals of varying breed composition, directly incorporates genomic data, produces true accuracy values and provides Stay EPDs to all IGS partners on a common base.
Features of the Multi-breed stayability with BOLT:
1) Multi-breed stayability. As mentioned, the new Stay EPD is calculated for all cattle in the database (not just for purebred and fullblood Simmental, as it was in past evaluations). The calculation of EPDs in this fashion required the new evaluation to account for heterosis, which can be sizable for a trait like stayability. Because stayability has a major impact on profit, having Stay EPDs on all cattle will substantially improve the ability of the All Purpose Index to predict differences in profit on hybrid and cattle of other breeds.
2) Direct incorporation of genomic data. This evaluation represents the first time "Single Step" methodology has been used in the beef industry on a large database. The innovative approach refers to DNA marker results being incorporated directly into the genetic evaluation, rather than being incorporated post evaluation or as a correlated trait. Single Step is widely accepted as the most powerful means of leveraging DNA for genetic prediction. The use of a Single Step approach on a large database (the largest in the beef industry) was made possible through the use of BOLT, ASA's revolutionary new software developed by researchers Dorian Garrick and Bruce Golden through Operation Quantum Leap.
3) More "accurate" prediction of accuracy. Prior to BOLT, the calculation of EPD accuracy was relegated to using indirect methods that were very imprecise, which commonly resulted in calculated accuracies being a poor estimate of the EPD's true accuracy, i.e., an EPD assigned a high accuracy may have truly been a low accuracy EPD and vice versa. Until the development of BOLT, the direct calculation of accuracy was thought to be impossible on a large database due to computational limitations. By leveraging statistical methodology unique to genetic evaluation and cutting-edge computer programming, BOLT calculates accuracy directly, thereby making it a much more "accurate" estimate of true accuracy.
Because the indirect methods used prior to BOLT tended to overestimate accuracy, you will notice that the accuracies associated with our new Stay EPDs will tend to be lower than the accuracies calculated in the prior evaluation of stayability. This does not mean that we are predicting stayability less accurately than we did in our previous evaluation--in fact, due to BOLT's ability to leverage phenotypes and genotypes more effectively we are actually predicting stayability with considerably more accuracy--it does mean that the published accuracy is now a better gauge of how much confidence you should have in an EPD. This fact will help seedstock and commercial producers better manage selection risk.
4) Change in base. Breeders will see significant movement in Stay EPDs from previous evaluations, not only because of improvements in methodology, but also because of a change in base. However, changing the base does not affect the ranking of cattle; it simply shifts all EPDs up or down by the same increment. For a point of reference, in the prior evaluation 21 was the average Stay EPD on purebred animals born over the last 2 years, where the average on the same animals is 11 under our new system.