Estimation & Prediction

A Multi-trait Single-step GBLUP Framework Integrating Ultrasound, Carcass, Genomic, and Imaging Data for Precision Genetic Evaluation in Hanwoo

---

Abstract

Hanwoo (Korean native cattle) is an economically important breed in Korea, and its breeding program has traditionally relied on pedigree-based genetic evaluation and carcass performance records collected post-slaughter. However, since carcass traits are measured only after slaughter, selection decisions for young and live animals are delayed, slowing genetic improvement. Ultrasound measurements have been adopted as indirect indicators of carcass quality, but variation in measurement conditions and evaluator subjectivity limit their reliability. This study aimed to establish a precision genetic evaluation system integrating ultrasound measurements, carcass data, genomic information, and ultrasound image features to improve selection accuracy in live Hanwoo cattle. Ultrasound data were collected from 2019 to 2024 using commercial devices. Among 6,193 animals, carcass phenotypes were available for 2,767, and genotypes for 2,696 animals genotyped using the Affymetrix Bovine 50K, Illumina Hanwoo 50K, and a customized Hanwoo 50K SNP array. Ultrasound traits (backfat thickness (U_BFT), eye muscle area (U_EMA), marbling score (U_MS)) and corresponding carcass traits (BFT, EMA, MS) were paired based on physiological and economic relevance. Genetic parameter estimation was performed using a linear mixed model. Multi-trait BLUP and Multi-trait single-step GBLUP were fitted to estimate genetic covariances between ultrasound and carcass traits. Heritability estimates for ultrasound traits ranged from 0.16 ± 0.02 (U_MS) to 0.26 ± 0.02 (U_BFT), and 0.35 ± 0.01 (EMA) to 0.46 ± 0.01 (MS) for carcass traits. Genetic correlations were high among ultrasound traits (r = 0.76-0.84) and moderate between ultrasound and carcass traits (r = 0.19-0.64), while phenotypic correlations were low (r = 0.02-0.38), indicating strong environmental influence on ultrasound measurements. For genomic prediction, Single-trait GBLUP with a reference population was applied to carcass traits, with the highest cross-validation accuracy for marbling score (MS; accuracy = 0.65). Additionally, an image-based classification model was developed using raw ultrasound images across six Hanwoo beef quality grades (1, 1+, 1++, 2, 3, D). The model employed the EfficientNet-B0 architecture trained with cross-entropy loss (label smoothing = 0.05) and optimized by AdamW (learning rate = 3à—10⁻⁴, weight decay = 1à—10⁻⁴). Images were resized to 384à—384 pixels, augmented, and trained for 15 epochs (batch size = 16) using 3-fold StratifiedGroupKFold validation. The best model achieved 30.6% accuracy and macro-F1 = 0.20, reflecting subtle visual variation and class imbalance. Nonetheless, the presence of informative image-derived signals suggests that image data can complement genomic and quantitative ultrasound information. In conclusion, this study establishes a unified evaluation framework from BLUP to Multi-trait and Single-step GBLUP, enabling integrated use of ultrasound, carcass, genomic, and imaging data to enhance live-animal selection accuracy and advance image-assisted genomic evaluation in the Hanwoo breeding program.

Keywords: 2026

How to Cite:

Lee, D., Go, J., No, C., Min, H., Koh, Y. & Lim, D., (2026) “A Multi-trait Single-step GBLUP Framework Integrating Ultrasound, Carcass, Genomic, and Imaging Data for Precision Genetic Evaluation in Hanwoo”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285379. doi: https://doi.org/10.31274/wcgalp.23667

Rights: 1