TKOA-GPNN: a model framework for genotype-to-phenotype prediction in Duroc pigs

Area of Science:

• Animal Genetics
• Bioinformatics
• Machine Learning in Animal Breeding

Background:

• Genomic prediction is crucial for pig breeding, utilizing large-scale genomic datasets.
• Neural networks excel at recognizing complex genetic patterns in high-dimensional data.

Purpose of the Study:

• Accurate prediction of pig growth traits is vital for genetic improvement due to their economic importance.
• Addressing the complexity of growth trait genetics and limitations of traditional neural networks in capturing genome-wide interactions.

Main Methods:

• Proposed TKOA-GPNN model with a two-stage genomic feature selection and a genomic prediction neural network.
• Utilized symmetric uncertainty correlation for feature space reduction.
• Employed Kepler optimization with tent mapping for trait-associated loci selection.

Main Results:

• TKOA-GPNN demonstrated superior accuracy and efficiency compared to 14 existing models.
• The model showed favorable performance with genotyping array data and robustness with whole-genome sequencing data.
• Confirmed applicability across different genomic data types and quality levels.

Conclusions:

• TKOA-GPNN significantly improves prediction accuracy and computational efficiency in pig genomic prediction.
• Offers an effective solution for intelligent pig breeding and genetic improvement.