Source: UNIV OF WISCONSIN submitted to
EXPLORING CASUAL RELATIONSHIPS UNDERLYING ECONOMICALLY IMPORTANT TRAITS IN DAIRY SHEEP
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
TERMINATED
Funding Source
HATCH
Reporting Frequency
Annual
Accession No.
1004763
Grant No.
(N/A)
Project No.
WIS01815
Proposal No.
(N/A)
Multistate No.
(N/A)
Program Code
(N/A)
Project Start Date
Oct 27, 2014
Project End Date
Sep 30, 2017
Grant Year
(N/A)
Project Director
ROSA, GU.
Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
Animal Sciences
Non Technical Summary
The U.S. is the world's largest importer of sheep milk cheese, accounting for about half of the world's sheep milk cheese in international commerce. However, for the national dairy sheep industry to continue to develop, further enhancement on its efficiency and competitiveness will be required. The success of livestock enterprises depends on a number of traits related to production and quality of animal products, and understanding the relationships among them is crucial for efficient decision making regarding farm practices and breeding strategies. For example, high yield may increase susceptibility to certain diseases and, conversely, the presence of disease may adversely affect yield. Hence, knowledge regarding the causal structure underlying phenotypic relationships is necessary to predict the effect of management practices applied to any single trait. In this project we will utilize specific statistical and data mining approaches to investigate causal relationships between production, reproduction and disease related traits in dairy sheep. We have extensive data available from the University of Wisconsin Spooner Station, in which records on a number of economically and socially important traits in the dairy sheep industry have been collected over the years in a commercial production environment. We hypothesize that the application of causal modeling approaches to the Spooner data will shed light on how various phenotypic traits are related to each other in terms of functional links, such that optimized management practices and genetic improvement of sheep populations can be developed with the consequent enhancement of dairy sheep production and sustainability. This research will contribute with data analysis tools for studying multiple traits in livestock production, as well as with fundamental knowledge regarding relationships among phenotypic traits in dairy sheep, which will contribute to optimized farm management practices for sustainable production systems.
Animal Health Component
0%
Research Effort Categories
Basic
40%
Applied
40%
Developmental
20%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
3073610106030%
3033610108030%
3073999106020%
3057310106020%
Knowledge Area
303 - Genetic Improvement of Animals; 305 - Animal Physiological Processes; 307 - Animal Management Systems;

Subject Of Investigation
3610 - Sheep, live animal; 3999 - Animal research, general; 7310 - Experimental design and statistical methods;

Field Of Science
1060 - Biology (whole systems); 1080 - Genetics;
Goals / Objectives
In this project, we will apply such methods for investigating causal links between economically important traits in sheep production, leveraging on extensive data available from the UW Spooner Station. It is our hypothesis that such application of causal modeling approaches to the Spooner data will shed light on how various phenotypic traits are related to each other in terms of functional (causal) links. Such knowledge can then be used for optimized management practices and genetic improvement of sheep populations, with the consequent enhancement of dairy sheep production and industry sustainability. In this context, the following specific aims will be developed:Specific Aims:1. Data analysis using traditional multiple trait models2. Application of path analysis based on prior biological knowledge; and3. Full data-driven search of causal structures.
Project Methods
This is a continuation of work done in WIS0 1716, where we applied specific data mining tools for investigating causal links between economically important traits in sheep production, leveraging on extensive data available from the University of Wisconsin Spooner Station. The ewe flock at the Spooner Agricultural Research Station has numbered approximately 300 ewes for the past 20 years. From 1996 through 2012, performance records have been collected on 10,256 lambs from 1,268 dams and 117 sires. Data collected on ewes included fertility (lambed or did not lamb), prolificacy (litter size), lactation length, milk yield, fat yield, protein yield, and somatic cell count. Data collected on lambs included birth weight, weaning weight, post-weaning weight, and death or disposal date. After a careful data-editing step to eliminate data inconsistencies and potential recording errors, records on a number of production, reproduction and disease traits will be first analyzed using a traditional multi-trait 'animal model' with the numerator relationship matrix constructed from pedigree information.In this project, mixed model equations will be used to estimate the fixed effects and to predict the random effects, and the restricted maximum likelihood approach will be used for variance components estimation. On a second step, structural equation models will be used for estimating and testing functional relationships among traits, which are often not revealed by standard linear models. Bayesian methods employing Markov Chain Monte Carlo (MCMC) tools will be used for fitting such models. In addition, specific algorithms will be employed for data-driven selection of causal structures, such as the Inductive Causation (IC) algorithm. The IC algorithm is based on queries about conditional independencies between variables to try to recover a causal structure or a class of equivalent causal structures (causal structures that result in joint probability distributions with the same conditional independence relationships) from the joint distribution of the data. As such, the input of the algorithm is a correlation matrix between observable variables, from which marginal and conditional dependencies can be evaluated, whereas the output is a partially oriented graph representing a class of equivalent causal structures, which generally results on an important constraint on the initial causal hypothesis space that could be used to fit the SEM. Partially-oriented graphs are graphs with directed and undirected edges. The latter represent symmetric direct relationships between pairs of variables, since they do not specify direction of causal relationship. On a final step, a SEM will be fitted with the selected causal structure so that the magnitude of causal effects can be inferred.

Progress 10/27/14 to 09/30/17

Outputs
Target Audience:Researchers and extensionists working in production and genetic improvement of sheep will benefit from this research. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project supported a graduate student (Ms. Vera Ferreira). In addition, the project has attracted international visiting scholars to the University of Wisconsin-Madison: Dr. Tatsuhiko Goto, from the Obihiro University of Agriculture and Veterinary Medicine in Japan; Dr. Nora Bello, from the Kansas Stat University; Dr. Angelina Fraga, Federal University of Alagoas, Brazil; Dr. Fernando Brito, from the Brazilian Agricultural Research Corporation (EMBRAPA), Brazil; and Ms. Katrin Topner, Technical University of Munich, Germany. Another training opportunity related to this project was the short-course "Introduction to Graphical Models with Applications in Genetics and Genomics", presented at Iowa State University, June 19-23, 2017. How have the results been disseminated to communities of interest?The project results have been disseminated to communities of interest via peer-reviewed scientific publications, a book chapter, and invited talks and seminars. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? The three specific research aims were fully developed, and the project produced an additional side research project related to the utilization of propensity score (PS) techniques for causal inference in livestock. A specific application in dairy sheep was published in the Journal of Dairy Science (Ferreira et al., J. Dairy Sci. 100:8443-8450, 2017), which inferred the effect of prolificacy on milk yield was inferred using a PS approach. Quite often the investigation of causal relationships among variables is performed using observational data, due to the lack of randomized experiments, for example in the case of the causal effect involving phenotypic traits. However, the association between traits may arise not only from the effect of one on another, but also confounding background factors. Propensity Score (PS) methods address this issue, correcting for confounding in the different levels of the causal variable which allows unbiased inference of marginal effects. Here the objective was to estimate the magnitude of the causal effect of prolificacy on MY using PS based on Matched Samples. The causal variable prolificacy was considered as two levels: single or multiple (2,3 and 4) lamb birth. The outcome MY represented the volume of milk produced in the whole lactation. 1,166 pairs of single/multiple lamb birth ewes with similar PS were formed. The matching process reduced major discrepancies in the distribution of prolificacy for each confounder variable indicating bias reduction, i.e. all covariates were deemed balanced after matching (standardized bias = 20%). Therefore, the causal effect was estimated as the average difference within a single pair. The effect of prolificacy on MY was significant () and equal to 20.52 L, with a simple matching estimator and 12.62, after correcting for the remaining biases. A core advantage of causal over probabilistic approaches is that they allow inference of how variables would react as a result of external interventions. Therefore, results imply that management practices that increase prolificacy would positively affect MY. Those results are important in the guidance of management practices at the farm level. At the same time, PS analysis may be a valuable tool for the inference of causal effects from farm data collections, especially given its low cost and the fact that it may be more capable of reflecting "real world" conditions. It could also be implemented as the preliminary evaluation or a hypothesis generator for future randomized trials.

Publications

  • Type: Journal Articles Status: Published Year Published: 2017 Citation: Ferreira, V. C., Thomas, D. L., Valente, B. D. and Rosa, G. J. M. Causal effect of prolificacy on milk yield in dairy sheep using propensity score. Journal of Dairy Science 100: 84438450, 2017.
  • Type: Journal Articles Status: Published Year Published: 2017 Citation: T�pner, K., Rosa, G. J. M., Gianola, D. and Sch�n, C.-C. Bayesian networks illustrate genomic and residual trait connections in maize (Zea mays L.). G3-Genes Genomes Genetics 7:2779-2789, 2017.
  • Type: Journal Articles Status: Published Year Published: 2017 Citation: Dadousis, C., Pegolo, S., Rosa, G. J. M., Gianola, D., Bittante, G., Cecchinato, A. Pathway-based genome-wide association analysis of milk coagulation properties, curd firmness, cheese yield, and curd nutrient recovery in dairy cattle. Journal of Dairy Science 100: 1223-1231, 2017.
  • Type: Journal Articles Status: Published Year Published: 2017 Citation: Castro, L. M., Rosa, G. J. M., Lopes, F. B., Regitano, L. C. A., Rosa, A. J. M. and Magnabosco, A. J. M. Genomewide association mapping and pathway analysis of meat tenderness in Polled Nellore cattle. J. Anim. Sci. 95:19451956, 2017.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Ferreira, V. C., Dorea, J. R. R. and Rosa, G. J. M. Big data analysis of beef production and quality: An example with the Brazilian cattle industry. Journal of Animal Science 95(4): 49-50, 2017.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Dorea, J. R. R. and Rosa, G. J. M. Mining farm- and animal-level data to optimize beef cattle production. Journal of Animal Science 95(4): 363, 2017.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Huang, X. Y., Elston, R. C., Rosa, G. J. M., Mayer, J., Ye, Z., Kitchner, T., Brilliant, M. H., Page, D. and Hebbring, S. J. Electronic health record: an untapped resource for family-based genetic epidemiologic research. Genetic Epidemiology 41(7): 660, 2017.


Progress 10/01/15 to 09/30/16

Outputs
Target Audience:Researchers and extensionists working in production and genetic improvement of sheep will benefit from this research. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project supported one year of a MS student (Ms. Vera Ferreira). In addition, the project has attracted international visiting scholars to the University of Wisconsin-Madison: Mr. Keiichi Inoue, from the Kyoto University in Japan; Mr. Renan Pinto, from the University of Sao Paulo in Brazil; Dr. Angelina Fraga, Federal University of Alagoas, Brazil; Dr. Fernando Brito, from the Brazilian Agricultural Research Corporation (EMBRAPA), Brazil; and Ms. Katrin Topner, Technical University of Munich, Germany. These students/scientists however were supported by their own institutions. Another training opportunity related to this project was the short-course "Introduction to Graphical Models with Applications in Genetics and Genomics", presented at the 62nd Brazilian-International Congress of Genetics - Caxambu, Brazil, Sept 11-14, 2016. How have the results been disseminated to communities of interest?The project results have been disseminated to communities of interest via peer-reviewed scientific publications, a book chapter, and invited talks and seminars. What do you plan to do during the next reporting period to accomplish the goals?For the next reporting period, we will continue working on the development and application of propensity score methods for the analyses of farm-recorded data from livestock, as well as network analyses based on graphical models.

Impacts
What was accomplished under these goals? During the reporting period, we worked on the estimation of the causal effect of prolificacy on milk yield (MY) in dairy sheep, using the methodology of propensity scores. Quite often the investigation of causal relationships among variables is performed using observational data due to the lack of randomized experiments, for example in the case of the causal effect involving phenotypic traits. However, the association between traits may arise not only from the effect of one on another, but also confounding background factors. Propensity Score (PS) methods address this issue, correcting for confounding in the different levels of the causal variable which allows unbiased inference of marginal effects. Here the objective was to estimate the magnitude of the causal effect of prolificacy on MY using PS based on Matched Samples. Data consisted of 4,319 records from 1,534 crossbred dairy ewes. Confounders were lactation order (1th, 2nd and 3th - 6th) and dairy breed composition (< .5, .5-.75 and >.75 of East Friesian or Lacaune). The causal variable prolificacy was considered as two levels: single or multiple (2,3 and 4) lamb birth. The outcome MY represented the volume of milk produced in the whole lactation. 1,166 pairs of single/multiple lamb birth ewes with similar PS were formed. The matching process reduced major discrepancies in the distribution of prolificacy for each confounder variable indicating bias reduction, i.e. all covariates were deemed balanced after matching (standardized bias = 20%). Therefore, the causal effect was estimated as the average difference within a single pair. The effect of prolificacy on MY was significant () and equal to 20.52 L, with a simple matching estimator and 12.62, after correcting for the remaining biases. A core advantage of causal over probabilistic approaches is that they allow inference of how variables would react as a result of external interventions. Therefore, results imply that management practices that increase prolificacy would positively affect MY. Those results are important in the guidance of management practices at the farm level. At the same time, PS analysis may be a valuable tool for the inference of causal effects from farm data collections, especially given its low cost and the fact that it may be more capable of reflecting "real world" conditions. It could also be implemented as the preliminary evaluation or a hypothesis generator for future randomized trials.

Publications

  • Type: Theses/Dissertations Status: Published Year Published: 2016 Citation: Ferreira, V. C. Application of propensity score to learn causal relationships involving genetically correlated traits. M.S. Dissertation, University of Wisconsin-Madison. Madison  WI, USA, 2016.
  • Type: Journal Articles Status: Published Year Published: 2016 Citation: Inoue, K., Valente, B. D., Shoji, N., Honda, T., Oyama, K. and Rosa, G. J. M. Inferring phenotypic causal structures among meat quality traits and the application of a structural equation model in Japanese Black cattle. Journal of Animal Science 94:4133-4142, 2016.
  • Type: Journal Articles Status: Published Year Published: 2016 Citation: Abdalla, E. A., Pe�agaricano, F., Byrem, T. M., Weigel, K. A. and Rosa, G. J. M. Genome-wide association mapping and pathway analysis of leukosis incidence in a US Holstein cattle population. Animal Genetics 47, 395-407, 2016.
  • Type: Book Chapters Status: Published Year Published: 2016 Citation: Rosa, G. J. M., Felipe, V. P. S. and Pe�agaricano, F. Applications of Graphical Models in Quantitative Genetics and Genomics. In: Systems Biology in Animal Production and Health, Volume 1. Kadarmideen, H. (Ed.) Springer, 2016.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2016 Citation: Ferreira, V. C., Valente, B. D. and Rosa, G. J. M. Accounting for genetic effects as confounders in propensity score analysis. In: 5th International Congress on Quantitative Genetics, Abstract 184, Madison-WI, June 12-17, 2016.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2016 Citation: Li, H., Wu, X.-L., Bauck, S., Thomas, D. L., Murphy, T. W. and Rosa, G. J. M. Association study of lactation performance in a dairy sheep population using classic GWAS and genomic models. In: 5th International Congress on Quantitative Genetics, Abstract 133, Madison-WI, June 12-17, 2016.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2016 Citation: Pinto, R. M., Valente, B. D., Gazel Filho, A. B., Leandro, R. A. and Rosa, G. J. M. Investigating causal relationships underlying phenotypic traits of two fruit species of the Sapotaceae family via graphical models. In: 5th International Congress on Quantitative Genetics, Abstract 148, Madison-WI, June 12-17, 2016.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2016 Citation: T�pner, K., Rosa, G. J. M., Gianola, D. and Sch�n, C.-C. Bayesian networks illustrate phenotypic and genomic trait connections in Maize. In: 5th International Congress on Quantitative Genetics, Abstract 47, Madison-WI, June 12-17, 2016.


Progress 10/27/14 to 09/30/15

Outputs
Target Audience:Researchers and extensionists working in production and genetic improvement of sheep will benefit from this research. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project suported one year of a MS student (Ms. Vera Ferreira). In addition, the project has attracted international visiting scholars to the University of Wisconsin-Madison: Mr. Keiichi Inoue, from the Kyoto University in Japan; Dr. Angelina Fraga, Federal University of Alagoas, Brazil; Dr. Fernando Brito, from the Brazilian Agricultural Research Corporation (EMBRAPA), Brazil; and Ms. Katrin Topner, Technical University of Munich, Germany. How have the results been disseminated to communities of interest?The project results have been disseminated to communities of interest via peer-reviewed scientific publications, and various talks at scientific meetings and seminars both in the US and abroad, including Italy, Germany and France. What do you plan to do during the next reporting period to accomplish the goals?For the next reporting period, we will continue working on the development and application of propensity score methods for the analyses of farm-recorded data from livestock and start implementing network analyses based on graphical models.

Impacts
What was accomplished under these goals? During the reporting period, we worked mostly on Specific Aim 1, with the estimation of breed and heterosis effects on survival of crossbred lambs using traditional multiple trait models. In summary, the results indicate that the proportion of individual retained heterosis was positively associated (P < 0.05) with lamb survival. The predicted increase in survival of F1 lambs compared to purebred lambs was +8.8 and +14.6%, respectively. Predicted survival of meat breed lambs and maternal breed lambs was greater (P ≤ 0.01) than Lacaune lambs. Predicted survival of East Friesian lambs was consistently lower (P ≤ 0.01) than meat breed and maternal breed lambs during all periods. The predicted survival of East Friesian lambs was numerically greater, but not significantly different from Lacaune lambs. There was a lower (P < 0.01) survival of females compared to males through 1 d of age (-5.6%), but females had higher (P < 0.01) survival than males in the other 2 periods (2 to 30 d = +3.3% and 2 to 60 d = +6.0%). Through 1 d of age, lambs of triplet and greater birth types had lower (P < 0.01) survival than single lambs (-6.2%), and lambs from 1-yr-old dams had lower (P < 0.01) survival than lambs from 2-yr-old dams (+4.5%). Estimates of heritability of lamb survival ranged from 0.03 to 0.14 depending on the period analyzed. As an overall conclusion, the increase in the proportion of individual retained heterosis was the most important genetic factor associated with increased lamb survival in this study. In addition, to this analysis using traditional multiple trait models, we started working also on a more formal causal inference approach using a statistical tool called propensity score (PS). The PS analysis was used to infer the effect of number of lambs born on ewe milk yield (MY). Preliminary results indicate that the estimated causal effect of the number of lambs born on MY was 20.52 L, se = 3.77 L, 95% confidence interval = 13.13-27.91 L. The meaning of this result is that ewes who gave birth to a single lamb would be expected to have their MY increased by 20.52 L, on average, if they had given birth to multiple lambs. This implies that management practices that increase the number of lambs born (superovulation, crossing with more prolific breeds, or selecting for more prolific ewes) are expected to cause an increase in the amount of milk produced.

Publications

  • Type: Journal Articles Status: Published Year Published: 2015 Citation: Hu, Y., Rosa, G. J. M. and Gianola, D. A GWAS assessment of the contribution of genomic imprinting to the variation of body mass index in mice. BMC Genomics 16: 576, 2015.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2015 Citation: Ferreira, V. C., Rosa, G. J. M., Berger, Y. M. and Thomas, D. L. Effects of breed and hybrid vigor on lamb survival. Proceedings of the 63rd Annual Spooner Sheep Day, Spooner Agricultural Research Station, University of Wisconsin-Madison, Spooner  WI, pp. 23-25, August 22, 2015.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2015 Citation: Ferreira, V. C., Thomas, D. L., Valente, B. D. and Rosa, G. J. M. Number of lambs born and milk production. Proceedings of the 63rd Annual Spooner Sheep Day, Spooner Agricultural Research Station, University of Wisconsin-Madison, Spooner  WI, pp. 55-57, August 22, 2015.
  • Type: Journal Articles Status: Published Year Published: 2015 Citation: Ferreira, V. C., Rosa, G. J. M., Berger, Y. M. and Thomas, D. L. Survival in crossbred lambs: Breed and heterosis effects. Journal of Animal Science 93: 912-919, 2015.
  • Type: Journal Articles Status: Published Year Published: 2015 Citation: Valente, B. D., Morota, G., Pe�agaricano, F., Gianola, D., Weigel, K. A. and Rosa, G. J. M. The causal meaning of genomic predictors and how it affects the construction and comparison of genome-enabled selection models. Genetics 200: 483-494, 2015.
  • Type: Journal Articles Status: Published Year Published: 2015 Citation: Felipe, V. P. S., Silva, M. A., Valente, B. D. and Rosa, G. J. M. Using multiple regression, Bayesian networks and artificial neural networks for prediction of total egg production in European quails based on earlier expressed phenotypes. Poultry Science 94: 772-780, 2015.