Testing AI Against Traditional Models for Predicting Cattle Feed Efficiency

Jenn Hoskins
18th March, 2024

Testing AI Against Traditional Models for Predicting Cattle Feed Efficiency

Key Findings

  • Study from São Paulo State University found new ways to predict cattle feed efficiency
  • Machine learning methods outperformed traditional genetic prediction methods
  • These methods could lead to more sustainable cattle breeding practices
In the realm of animal breeding, particularly within the cattle industry, the concept of feed efficiency (FE) has become increasingly important. FE refers to how effectively an animal converts feed into body mass. Improving FE is not only economically beneficial for farmers, but it also has environmental implications, as more efficient animals require less feed and produce fewer emissions per unit of product. A recent study from São Paulo State University (UNESP)[1] has made significant strides in predicting FE-related traits in Nellore cattle, a breed primarily raised for beef in Brazil. The researchers compared several genetic prediction methods to determine which could most accurately forecast these traits, which are critical for selecting the most efficient animals for breeding. Traditionally, genomic selection (GS) has relied on methods like single-trait genomic best linear unbiased prediction (STGBLUP) and multi-trait genomic best linear unbiased prediction (MTGBLUP), as well as various Bayesian regression approaches. These methods use genetic markers spread across the genome to predict the breeding value of an animal for certain traits. However, the study revealed that machine learning techniques, specifically multi-layer neural network (MLNN) and support vector regression (SVR), outperformed these traditional methods. The researchers evaluated the accuracy of these methods using a dataset of approximately 300,000 genetic markers from 1156 Nellore cattle. They employed a forward validation approach, which means they tested the accuracy of their predictions on newer animals that weren't part of the initial training set. This is akin to using past data to predict future outcomes, ensuring the models are robust and can handle new information. The findings were clear: MLNN and SVR increased prediction accuracy by 8.9% and 14.6%, respectively, compared to STGBLUP. MTGBLUP also showed an improvement, with a 13.7% increase in accuracy. SVR and MTGBLUP were almost neck and neck in their performance, with accuracies ranging from 0.62 to 0.69 and 0.62 to 0.68, respectively. These results are significant because they suggest that machine learning methods could be more suitable for genomic selection in cattle, especially for complex traits like FE that have a multifaceted genetic basis. This could lead to more rapid genetic progress and, ultimately, more sustainable cattle production. The study builds on previous research that has explored different aspects of genomic prediction. For instance, a guide on implementing Bayesian generalized kernel regression methods for genomic prediction[2] provided a foundation for capturing complex non-linear patterns in genetic data, which is something machine learning models excel at. Additionally, earlier work on optimizing weighted genomic relationship matrices for genomic prediction[3] highlighted the importance of considering the genetic architecture of traits, which is also a consideration in machine learning approaches. In the context of environmental sustainability, the use of correlated traits to increase the accuracy of genomic estimated breeding values (GEBV) for methane emissions in dairy cattle[4] is analogous to the multi-trait approach in the current study. Both strategies leverage additional information to enhance prediction accuracy for traits of interest. Lastly, the identification of candidate genes and pathways related to feed efficiency in Nellore cattle[5] complements the predictive approaches by potentially providing biological insights that could further refine genomic selection strategies. In conclusion, the study from UNESP demonstrates that machine learning methods, such as MLNN and SVR, and the MTGBLUP approach, are promising tools for improving the prediction of FE-related traits in cattle. These methods could revolutionize genomic selection by providing more accurate and efficient ways to select for economically and environmentally favorable traits in livestock. As the industry continues to evolve, integrating these advanced predictive models could be key to meeting the dual challenges of profitability and sustainability.

BiotechGeneticsAnimal Science

References

Main Study

1) Benchmarking machine learning and parametric methods for genomic prediction of feed efficiency-related traits in Nellore cattle.

Published 17th March, 2024

https://doi.org/10.1038/s41598-024-57234-4

Related Studies

2) A guide for kernel generalized regression methods for genomic-enabled prediction.

https://doi.org/10.1038/s41437-021-00412-1

3) Efficient weighting methods for genomic best linear-unbiased prediction (BLUP) adapted to the genetic architectures of quantitative traits.

https://doi.org/10.1038/s41437-020-00372-y

4) Multitrait genomic prediction of methane emissions in Danish Holstein cattle.

https://doi.org/10.3168/jds.2019-17857

5) Weighted single-step genome-wide association study and pathway analyses for feed efficiency traits in Nellore cattle.

https://doi.org/10.1111/jbg.12496


Related Articles

An unhandled error has occurred. Reload 🗙