Enhancing Genomic Prediction in Dairy Cows with Advanced Neural Networks

Jim Crocker
2nd July, 2024

Enhancing Genomic Prediction in Dairy Cows with Advanced Neural Networks

Image Source: Ann H (photographer)

Key Findings

  • Researchers at China Agricultural University developed Biologically Annotated Neural Networks (BANNs) for genomic prediction in dairy cattle
  • BANNs outperformed traditional methods like GBLUP and Bayesian approaches in predicting genetic traits
  • BANNs offer better accuracy and interpretability by incorporating biological knowledge and capturing complex genetic interactions
Genomic prediction, the process of predicting the genetic value of an individual based on their DNA sequence, has seen significant advancements with the advent of high-throughput sequencing and sophisticated statistical models. One recent development in this field is the introduction of Biologically Annotated Neural Networks (BANNs) by researchers at China Agricultural University[1]. BANNs are a type of Bayesian neural network designed to incorporate biological knowledge into their architecture, making them both powerful and interpretable. Traditional genomic prediction methods, such as Best Linear Unbiased Prediction (BLUP) and its genomic counterpart (GBLUP), have been widely used to predict genetic values based on dense marker maps[2]. These methods assume simple genetic architectures and equal variances across chromosomal segments, which can lead to inaccuracies. More advanced methods, like Bayesian approaches, have improved prediction accuracy by incorporating prior distributions of genetic variances[2]. However, these methods often overlook the complex interactions between genetic variants. The implementation of genomic selection in dairy cattle breeding has demonstrated the potential of genomic prediction to accelerate genetic gains[3]. For instance, the use of genomic evaluations in French dairy cattle breeds has significantly increased annual genetic gains while reducing generation intervals. However, this has also led to concerns about increased inbreeding, particularly in the Holstein breed, highlighting the need for more sophisticated models that can balance genetic gain with genetic diversity. BANNs aim to address some of these limitations by using partially connected architectures based on SNP-set (single nucleotide polymorphism) annotations. In a BANN, SNP and SNP-set effects are modeled in the input and hidden layers, respectively, allowing the network to capture complex genetic interactions and hierarchical genetic effects. The weights and connections in the network are treated as random variables with prior distributions, reflecting the manifestation of genetic effects at various genomic scales. This approach contrasts with traditional linear mixed models (LMMs), which typically assume simple genetic architectures and often fail to capture local genetic interactions. Recent extensions of LMMs, such as the multikernel linear mixed model (MKLMM), have improved the modeling of genetic interactions by incorporating multiple kernels that can represent different types of genetic relationships[4]. MKLMM-Adapt, an extension of MKLMM, automatically infers interaction types across genomic regions, further enhancing prediction accuracy and computational efficiency[4]. In addition to improving prediction accuracy, BANNs also offer interpretability, a crucial feature for understanding the biological basis of complex traits. This interpretability is achieved by structuring the network based on biological annotations, such as SNP-sets, which group SNPs based on their functional or positional relationships. This allows researchers to gain insights into the genetic architecture of traits and identify key genetic regions that contribute to trait variation. Comparing BANNs with other advanced genomic prediction methods, such as the Cosine kernel-based KRR (KCRR), highlights the potential advantages of BANNs. KCRR has shown stable performance across various traits and species by using a modified genomic similarity matrix, significantly improving computational efficiency[5]. However, while KCRR excels in computational efficiency, BANNs offer the added benefit of biological interpretability, making them a valuable tool for both prediction and understanding of genetic mechanisms. In summary, the development of Biologically Annotated Neural Networks by researchers at China Agricultural University represents a significant advancement in genomic prediction. By incorporating biological knowledge into the neural network architecture, BANNs can model complex genetic interactions and provide interpretable insights into the genetic basis of traits. This approach builds on and extends previous methods, offering a promising strategy for improving prediction accuracy and understanding genetic diversity in breeding programs.

BiotechGeneticsAnimal Science

References

Main Study

1) Improving the accuracy of genomic prediction in dairy cattle using the biologically annotated neural networks framework

Published 1st July, 2024

https://doi.org/10.1186/s40104-024-01044-1

Related Studies

2) Prediction of total genetic value using genome-wide dense marker maps.

Journal: Genetics, Issue: Vol 157, Issue 4, Apr 2001

3) The impact of genomic selection on genetic diversity and genetic gain in three French dairy cattle breeds.

https://doi.org/10.1186/s12711-019-0495-1

4) Multikernel linear mixed models for complex phenotype prediction.

https://doi.org/10.1101/gr.201996.115

5) KCRR: a nonlinear machine learning with a modified genomic similarity matrix improved the genomic prediction efficiency.

https://doi.org/10.1093/bib/bbab132


Related Articles

An unhandled error has occurred. Reload 🗙