Urine, saliva, and serum metabolomics predict female pig reproductive potential status using machine learning modelling.

Profitability in the pork industry relies heavily on breeding herd productivity. However, traditional phenotypic and genetic methods are limited in their selection efficiency for some reproductive potential related traits. This leads to high culling rates due to poor reproductive performance and detriments to productivity and profits. This study investigated whether metabolomic profiling of urine, saliva, and serum applied to machine learning (ML) algorithms could predict female pig reproductive potential status. Urine, saliva, and serum samples from high reproductive potential (HRP; lifetime number of piglets born alive ≥ 13) or infertile (INF; nonpregnant after two consecutive rounds of artificial insemination) female pigs were analyzed using targeted liquid chromatography and mass spectrometry. Metabolomic profiles were applied to an ML pipeline including partial least squares discriminant analysis and recursive feature elimination complementary feature selection and six supervised ML algorithms. The urine-based logistic regression (F1 = 0.90 ± 0.03, Matthews correlation coefficient = 0.81 ± 0.11) and saliva-based logistic regression (F1 = 0.94 ± 0.04, Matthews correlation coefficient = 0.86 ± 0.10) classifiers achieved impressive predictive performance, demonstrating the potential to accurately measure female pig reproductive potential. Feature importance and univariate analysis highlighted key biomarkers, including specific carnitines, amino acids, and phospholipids, that distinguished the female pig reproductive metabolic phenotype. Overall, integrating urine, saliva, and serum metabolomic data with ML offers a promising approach to improve female pig reproductive potential prediction. This could assist breeding herd sow retention and culling protocols, improving production and economic outcomes. Future work should validate these classifiers for practical use by including additional samples from broader farming environments and assessing the feasibility of integrating this approach in commercial operations.