# Verification of Accuracy of Genomically Enhanced Predicted Transmitting Ability Techniques in Predicting Milk and Fat Production in Holstein Cattle in Taiwan

## TL;DR

This study shows that genetic predictions can help estimate milk and fat production in Holstein cows in Taiwan, but environmental factors also play a big role.

## Contribution

The study demonstrates the predictive accuracy of genomic evaluations in subtropical dairy systems and identifies trait-specific environmental interactions.

## Key findings

- Genomic predictions for milk and fat yield showed moderate accuracy (R2 up to 0.507) in subtropical conditions.
- Environmental factors like feed and climate significantly affect fat yield but not milk production.
- Incorporating herd and birth-year effects improved prediction accuracy and model robustness.

## Abstract

This study examined how well genetic information can predict milk and fat production in Holstein dairy cows in Taiwan. By analyzing records from 986 cows across 25 farms, we found that cows with higher genetic scores for milk and fat production generally produced more milk and butterfat in their first lactation. When farm and birth-year effects were considered, prediction accuracy improved substantially (R2 increased from 0.12 to 0.47), emphasizing that environmental and management factors strongly influence gene expression in subtropical herds. The results also showed that milk production was more stable across farms, while fat yield was more sensitive to environmental conditions such as feed and climate. These findings highlight the value of integrating genomic evaluations with farm-level breeding strategies to enhance productivity and genetic progress in subtropical dairy systems like Taiwan.

This study evaluated the predictive performance of genomically enhanced predicted transmitting abilities for milk (gPTAM) and fat yield (gPTAF) in 986 first-lactation Holstein cows from 25 herds in Taiwan. Ordinary least squares and linear mixed models revealed significant positive associations between genomic predictions and observed yields (milk: β = 1.201, R2 = 0.469; fat: β = 1.444, R2 = 0.507). Incorporating herd and birth-year effects improved model fit and reduced residual variability. Five-fold cross-validation confirmed the robustness of the full mixed model, with predictive R2 increasing to 0.293 for milk and 0.363 for fat, distinct from the OLS R2 (0.469 and 0.507) representing phenotypic variation explained, indicating moderate predictive ability of genomic PTA values under subtropical production conditions. Model adequacy checks supported appropriate model specification, with only a mild outlier signal in the milk model. Regional analysis revealed a significant genotype-by-environment interaction for PTAF (p = 0.018) but not for PTAM, indicating reduced prediction accuracy in environmentally variable regions and highlighting trait-specific environmental sensitivity. Quartile stratification by gPTA and Net Merit Score demonstrated clear yield gradients, confirming both the predictive and economic value of genomic evaluations under subtropical dairy production systems.