Hanwoo is the most popular beef cattle breed in Korea and is prized for its excellent marbling, tenderness, juiciness, and unique flavour. It is a native taurine breed of Korea that was historically used as a draft animal, dating back 5,000 years [1]. Traditionally, it has been utilized in farming, transportation, and religious ceremonies, but it has undergone a significant transformation over the past 30 years into a specialized meat-type breed [2]. This shift was driven by the rapid economic growth of Korea in the 1960s, which increased beef consumption and supported commercial Hanwoo production [1].

Korean consumers have valued Hanwoo beef for its flavour, tenderness, marbling, and health benefits, including low cholesterol and high omega-3 content, which are associated with a lower risk of cardiovascular disease [3], offering a healthier alternative than other beef breeds. As a result, this economically important breed has become the focus of extensive research aimed at improving both meat quantity (protein content) and quality (taste and health consciousness of meat). Thus, carcass traits, including carcass weight (CWT), backfat thickness (BFT), the eye muscle area (EMA), the marbling score (MS), and the rib eye area (REA), have been prioritized as selection criteria in commercial Hanwoo breeding programs. Since the 1960s, HC has undergone remarkable genetic improvements, resulting in exceptional enhancements in meat quality, which makes it one of the world’s leading sources of premium, highly marbled beef [4]. Early genetic improvement efforts focused mainly on carcass quality and growth traits because of the availability of trait information and simple analytical tools [5]. Figure 1 illustrates the major research themes and keywords from the literature on HC, highlighting the long-term scientific focus on meat quality and carcass improvement. Factually, carcass traits (CWT, BFT, EMA, and MS) are quantitative traits that are moderately to highly heritable [6,7] and polygenic in nature and are controlled by multiple genes with small-effect SNPs [8,9]. Despite their economic importance, owing to their direct influence on meat yield, carcass quality grade, and overall profitability, these traits are difficult to measure in live animals [10], thereby emphasizing the necessity of GS to achieve more accurate and efficient genetic improvement.

Historically, breeding programs relied on pedigree and phenotypic information to identify superior animals. The estimation of breeding values (EBVs) using the best linear unbiased prediction (BLUP) method developed by Henderson [11] represented a fundamental step in genetic evaluation. When direct phenotyping of selection candidates is impossible (such as with lethal traits or sex-linked traits), breeding values are traditionally predicted via data from relatives (full or half-sibs) or pedigree records. Without genomic information, traditional methods provide the same breeding value to sibs, ignoring within-family genetic differences. Interestingly, the advent of GS addressed this limitation by utilizing genome-wide SNP information to distinguish individual genetic differences. Consequently, GS has since become a cornerstone of modern animal breeding, offering enhanced accuracy, efficiency, and genetic gain across species and traits by leveraging dense genomic information [12]. Moreover, GS enables a significant decrease in the generation interval by accelerating the accuracy of breeding in young animals [13]. With the introduction of high-density SNP panels and thousands of identified markers, the prediction of genetic merit through genomic estimated breeding values (GEBVs) became more reliable [14]. Moreover, GS could increase accuracy by reducing the need for phenotypes for different species when traits are difficult or costly to measure with low heritability [15] or strong genotype-environment interactions [16]. Initially, the adoption of GS was limited by the high cost of genotyping; however, as costs continue to decrease, its use has expanded [17]. On the other hand, fully replacing phenotyping with genotyping shortens the breeding cycle by enabling early selection without waiting for phenotypic data [18]; partial phenotyping replacement filters out poor candidates, leading to increased selection intensity [19]; and finally, the remaining candidates are evaluated with both genomic and pedigree information at no extra cost [17]. Furthermore, genomic information helps to select parents with high breeding values with complementary traits, resulting in improved progeny quality and providing more precise genomic relationships than pedigree relationships, aiding in maintaining genetic diversity. Therefore, interest in the use of GP methods has also increased among the beef industry. Importantly, various genomic evaluation approaches have shown similar levels of prediction accuracy [20] but with higher accuracies than the traditional pedigree-based best linear unbiased prediction (PBLUP) [21,22].

In Hanwoo, recent efforts have focused on implementing GP methods to improve breeding programs and enhance economically important traits, particularly carcass traits, which are difficult to measure in live animals. Research contributions from 1970 to 2025 have demonstrated the rapid development of statistical methods in Hanwoo breeding (Figure 2). However, studies found positive genetic correlations among Hanwoo carcass traits that enable simultaneous improvement, making them ideal for GS-based selection [23,24]. The genomic models that utilize SNP information can effectively capture Mendelian sampling effects [25], which helps make breeding predictions more accurate. Furthermore, the integration of high-density SNP arrays and multiomics data has contributed to more accurate GEBVs for key carcass traits [8,23,26,27]. Empirical evidence supports the effectiveness of GS in Hanwoo; for example, Seo et al [28] reported annual genetic gains of 0.78 kg in yearling weight, 0.35 kg in CWT, and 0.27 cm² in the EMA. Moreover, between 2000 and 2016, the average CWT of Hanwoo steers rose from 343 kg to 437 kg, alongside an improvement in MS from 3.6 to 5.6 [29], reflecting substantial progress in meat quality. This continued genetic progress encourages Hanwoo breeders to place greater emphasis on the adoption and expansion of GS strategies. Despite significant progress, there are still several challenges observed in the application of GS to the Hanwoo breeding program. The limited size of the Hanwoo reference population continues to constrain the accuracy of genomic evaluation. In addition, integrating multi-omics data (such as transcriptomics, metabolomics, and epigenomics) into breeding programs is still underdeveloped. Researchers also face difficulties in applying advanced statistical and computational methods, particularly machine learning (ML), due to data complexity, heterogeneous trait architecture, and the need for large, well-structured datasets. Addressing these gaps is crucial for maximizing the benefits of genomic advances in Hanwoo breeding. The overall framework of the GS procedure, encompassing reference population establishment, genotyping, prediction og GEBVs, and selection of superior breeding animals, is illustrated in Figure 3. In recent years, genomic best linear unbiased prediction (GBLUP) and Bayesian approaches have been widely adopted to increase the prediction accuracy of Hanwoo over PBLUP [30–33]. Moreover, ML has also emerged as a promising approach in the genomic era, because of their tremendous flexibility and ability to model complex and nonlinear patterns in high-dimensional and noisy genomic datasets [34], where linear models most often face statistical challenges and show poor performance [35,36]. For example, breeders want to increase MSs; in such cases, the ML model, particularly deep learning, could be a highly promising prediction model because it could handle the challenges in Marbling [37].

Overall, this review synthesizes the current advances in GP models applied to Hanwoo cattle, highlighting the key milestones in breeding strategies and improvements in the predictive accuracy of carcass traits. Our study underscores the need for continued development and refinement of GS programs to achieve sustainable genetic progress by enhancing economically important traits while preserving superior meat quality. GS remains the most powerful and efficient framework to achieve these objectives, enabling accelerated genetic gain through accurate estimation of genomic breeding values. Moving forward, expanding reference populations, integrating multi-omics data, and optimizing ML algorithms will be indispensable for further improving the accuracy, robustness, and long-term sustainability of genomic evaluation systems in Hanwoo beef cattle.

Several animal-based statistical models and ML algorithms have been developed with the introduction of the GS concept over the traditional method to strengthen genetic gains by selecting superior animals early in life or even before birth. However, an essential fact in GP is the selection of a suitable statistical method, which is reliant upon trait complexity, population structure, and resource processing. The fundamental disparities among the statistical models that have been put forth thus far are found in the assumptions that they make about the distributions of SNP effects and the genetic architecture underlying the traits [38]. In addition, a major challenge arises when the number of markers greatly exceeds the number of observations, leading to limited degrees of freedom and more variables than with individuals [39]. Notably, such challenges could be addressed by using SNP-BLUP, GBLUP, Bayesian regression models, and ML algorithms. However, the prediction through EBVs and GEBVs in Hanwoo cattle is factually coordinated by the National Institute of Animal Science (NIAS) under the Rural Development Administration (RDA), Korea. NIAS operates the Hanwoo progeny testing and improvement program, which collects performance, pedigree, and genomic data from nationwide herds. The system provides a structured framework for the selection of superior artificial insemination (AI) bulls and replacement heifers based on genetic merit. Moreover, EBVs or GEBVs estimation for carcass and growth traits forms the foundation of the official Hanwoo selection index used in national breeding programs. The estimation of the breeding value of HC via different methods is illustrated in Figure 4. Some of the most common approaches applied in the Hanwoo breeding program, along with traditional methods, are briefly discussed in the following sections:

Figure 6 depicts the key milestones of the Korean HC breeding strategy [1,4]. In the 1960s, with the rapid growth of the Korean economy, meat demand surged, prompting the government to establish the Livestock Industry Act to guide Hanwoo improvement as a beef breed. In 1969, the first National Hanwoo Championship was held, and three proven bulls were selected for a nationwide AI program. During the 1970s, farmers’ strong interest led to efforts focused on increasing body weight through both crossbreeding and pure breeding. However, crossbreeding has failed to improve beef quality or reduce production costs compared with imports. As a result, it was discontinued, and the focus shifted entirely to pure breeding, emphasizing both qualitative and quantitative improvements in meat traits. In 1979, the first national breeding program, the Hanwoo Improvement Complex (HIC), or Hanwoo Gaeryang Danji, was launched by the Ministry of Agriculture and Forestry (MAF) to promote purebred cow-calf production. Nevertheless, HIC remained unimpressive and was stopped in 1998 because of limited farmer involvement, poor record-keeping, and a shortage of skilled staff. In 1983, cow-calf farms were established, and the pure-breeding policy was widely adopted. As a follow-up initiative, the Hanwoo proven bull selection program was introduced in two stages: i) a performance test (PT) in 1983 and ii) a progeny test (PGT) in 1985 to improve the body growth rate and meat quality. This led to the first selection of 10 proven bulls in 1987. The Hanwoo improvement farm (HIF), or Hanwoo Gaeryang Nongga, was established in 1993, similar to the HIC, but focused on individual farm evaluations rather than a group of farms together. After a successful 5-year pilot experiment, HIC was replaced, and HIC was continued until 2009. However, short breeding windows, unstructured mating plans and dam selection, minimal cow participation, and inadequate emphasis on cow performance beyond pedigrees and appearance made it difficult for both the HIC and HIF programs to successfully support the Hanwoo proven bull program. Consequently, the MAF introduced a major amendment in 2004 to improve cow-calf farm management, with a particular focus on performance recording, seasonal breeding, and disease control. Following this great initiative, the new Hanwoo Nucleus Breeding Farm program, or Hanwoo Yookjong Nongga, started in 2005 and achieved significant success with the establishment of 104 nucleus farms. In 2010, after HIF ended, the MAF launched the Progressive Farmers’ Group Support program. It supported farmers who invested in livestock improvement and required payments for specific services (ultrasound scans, animal registration, and breeding consultation). A major improvement was the inclusion of comprehensive consulting services, including genetic evaluation and mating plans, offered by 11 expert institutions authorized by NIAS, ensuring consistency and quality, an area where the HIF was lacking. In Hanwoo, pedigree-based selection achieved gradual improvement in breeding objective traits such as CWT and EMA with a slight negative genetic response (−0.036) of MS per year [1,41]. Besides, the prediction accuracy of EBVs is low to moderate (Table 1), which largely depends on pedigree depth, record quality, and herd size [41]. However, traditional breeding methods are often inaccurate and insufficient involving, long generation intervals, slow genetic gains, and low prediction accuracy, particularly for low-heritability traits such as MS. These limitations were addressed with the advent of genome-wide SNP panels, leading to the development of a reference population with high-density SNP genotypes between 2012 and 2013. Despite this, NIAS and RDA developed the Hanwoo 50K SNP chip for the application of genomic selection to HC. The first GP models were subsequently trained on reference population data from 2015 and subsequently applied to the carcass traits of Hanwoo. Since 2016, broader GS adoption in the Hanwoo breeding program has started. This transition of Hanwoo to GS aimed to increase prediction accuracy, accelerate genetic gain, and enable broader utilization of nationwide carcass data, thereby supporting the ongoing Hanwoo Performance and Progeny testing programs.

For HCs, whose genetic improvement over the past 40 years has been heavily focused on meat productivity [64], GS presents a strategic opportunity to expand selection goals and accelerate progress. The GP accuracies of the carcass traits of the Korean Hanwoo beef cattle are summarized in Table 1. However, the accuracy of GP using 50K SNP chips has shown variable performance depending on the reference population size, trait architecture, and marker density [1]. Table 2 presents a summary of the current investigations on GS using various GP methods and different genomic data for carcass traits in Hanwoo beef cattle.

The use of ssGBLUP, which integrates genotyped and non-genotyped individuals through joint modeling of pedigrees and genomic information, has become the gold standard in GS implementation [65]. In Hanwoo, this approach has been adopted in the national evaluation system following initial use of a multistep GBLUP model, leading to significantly improved predictive performance for carcass and body measurement traits [23,41,47]. Despite challenges in beef cattle, such as incomplete pedigrees, small progeny groups, and strong maternal effects [66], ssGBLUP has outperformed PBLUP in breeds such as Japanese Black, Angus, and Nellore [67–69]. Additionally, multi-trait models further increase prediction accuracy, particularly for correlated traits, as demonstrated by Jang et al [26], who reported higher prediction accuracy using a multi-trait ssGBLUP model for MS than both GBLUP and BayesR.WGBLUP and WssGBLUP have been developed to account for differential SNP effects. WGBLUP with nonlinear SNP weighting significantly increased prediction accuracy by 4.84% (BFT) and 2.70% (CWT) beyond that of GBLUP, but it has no added advantage over GBLUP for EMA and MS in HCs [49]. Similarly, when SNP weights and information from both genotyped and non-genotyped animals are incorporated, window-based weighting (WssGBLUP) significantly enhances accuracy, particularly for CWT (22%) and EMA (15%), with no accuracy gain for BFT and MS, indicating that these traits are influenced by major QTLs [53]. Moreover, the WssGBLUP model could outperform GBLUP for traits whose QTLs have moderate to large impacts [70]. Therefore, it is worth mentioning that WGBLUP or WssGBLUP is most beneficial for traits with large-effect QTLs, making it a trait-specific tool for increasing GP. Another study [51] reported that WGBLUP outperforms traditional GBLUP, where significant SNPs with distinct weights were considered to predict the accuracy of both the reproductive and carcass traits of HC. In addition, this study clarifies how important it is to include heterogeneity in SNP effect sizes in genomic assessments to increase prediction accuracy.

In HC, BayesR yielded higher prediction accuracies for CWT and BFT when the 50K SNP array was used [8]. BayesR also facilitated the identification of SNPs with substantial effects (ranging from 4.02% to 6.92%) on carcass traits, particularly within chromosomal regions BTA6 and BTA14, which are known to harbor influential QTLs [6]. This enhancement is attributed to the model’s ability to assign differential SNP effects, thereby more effectively capturing the influence of major QTLs. In comparative evaluations, both BayesR and GBLUP outperformed the traditional PBLUP, reinforcing the suitability of genomic models for selection in Hanwoo [30]. BayesC also outperformed both GBLUP and Bayesian LASSO for CWT, providing a 7% increase in accuracy, whereas performance for BFT, EMA, and MS remained comparable across methods [60]. These results suggest that CWT is regulated by a limited number of loci with large effects, whereas other carcass traits are more polygenic in nature. Further supporting evidence is provided by Lee et al [52], who applied a single-step Bayesian approach to a dataset comprising 988 genotyped and 1,438 non-genotyped Hanwoo individuals. This method yielded improved accuracy for CWT and demonstrated robust performance across all four carcass traits, underscoring its efficacy in integrating genomic and pedigree information. BayesB also produced higher heritability estimates for CWT and EMA, whereas GBLUP yielded better estimates for MS [33], suggesting differential model suitability depending on trait architecture. Similarly, BayesC outperformed GBLUP and Bayesian LASSO for CWT with a 7% gain, indicating the oligogenic nature of this trait [60]. Additionally, Bayesian models (e.g., MT-BayesCπ and mbBayesAB) have shown superior prediction accuracy over GBLUP for low-heritability reproductive traits in beef cattle [71,72] as well as in Hanwoo cows [73], highlighting the utility of Bayesian methods for traits influenced by major QTLs or with low heritability.

Recent studies have highlighted the increasing role of ML and deep learning in GP. In Chinese Simmental, ML algorithms such as SVR, KRR, RF, and AdaBoost. RT outperformed GBLUP for predicting the GEBVs of CWT, EMA, and live weight [74], whereas in Hanwoo, techniques such as MLR, NN, and PLSR were used to predict carcass traits from body measurements [27]. Similarly, Shahinfar et al [62], using data from 2,108 HCs, reported strong performance of ML methods for predicting MS and carcass traits, with SVM-SMO (support vector machine with sequential minimal optimization) achieving the best results. Table 3 summarizes different algorithms of ML and deep learning for the Hanwoo population. However, ML models do not always outperform traditional methods. For example, Srivastava et al [7] reported no consistent ML advantage over GBLUP, although XGBoost achieved higher accuracy for CWT and MS. These inconsistencies highlight that the performance of the ML model is highly context- and trait-dependent. Moreover, recent studies [10,75] have demonstrated the potential of deep learning for precision livestock management, achieving over 97% accuracy in identifying individual calves and HCs via muzzle pattern recognition, indicating strong applicability for real farms.

Furthermore, the integration of computer vision and digital phenotyping in GSs is an emerging field. Recently, Lee et al [76] introduced the F7 index, a novel marbling fineness score, which is based on the standard deviation of the marbling particle area across a grid and shows strong discriminatory power between BMSs 6–9 (p = 1.34×10⁻²⁷). Such trait innovations can be integrated into GS programs, allowing breeders to target premium quality attributes more precisely. Another interesting study [54] in which the association of a genome-wide approach and GWAS resulted in a GP (called GWABLUP) yielded up to 10% and 13% more reliable GPs than did GBLUP for single and multi-trait analyses, respectively.

However, Hanwoo GP has largely focused on carcass and meat quality traits; similar trends have also been reported in other beef cattle. For instance, in Japanese black cattle, Onogi et al [67] observed higher predictive accuracies using ssGBLUP, particularly for CWT (0.84) and EMA (0.79), compared to Hanwoo [31]. Moreover, Ogawa et al [77] observed accuracies for CWT (0.85) and MS (0.60) using GBLUP with G matrix under different SNP densities, which are comparable with [60]. In contrast, the tropical beef breeds such as Nellore showed a lower predictive accuracy using GBLUP (0.30–0.38) and Bayesian methods (0.31–0.42) for the carcass traits (EMA, BFT, CWT) due to higher genetic diversity and limited reference populations [20]. Another study by Wang et al [78] investigated the prediction accuracy of GEBVs for growth, carcass, and meat quality traits in Chinese Simmental cattle, using BayesB and Elastic Net models, where the prediction accuracy (0.17 to 0.296) was comparatively lower than other beef breeds. In Angus, the high level of relatedness in training and validation animals substantially enhanced prediction accuracies for a range of carcass and production traits, ranging from 0.22–0.69 under low-relatedness to 0.38–0.85 under random splits with higher relatedness [79]. Similarly, in Hanwoo, low relatedness between reference and validation animals resulted in noticeably reduced accuracy for carcass traits, ranging from 0.21 to 0.45 [60]. Nevertheless, these comparisons indicate that advancements in Hanwoo genomic evaluation align well with global trends, while accuracy could be further enhanced by expanding reference populations.

The success of GS depends on the ability to accurately predict GEBVs across generations, ideally without the need for repeated phenotyping once marker effects have been estimated. Importantly, the accuracy of GEBVs relies on reference population size, trait heritability, linkage disequilibrium, and QTLs [80]. Despite its proven utility in dairy cattle, the adoption and impact of GS in beef cattle, including HCs, have been relatively limited [81]. However, the adoption of GS in HC breeding offers substantial advantages in terms of prediction accuracy, reduced generation intervals, and genetic gains. It enables early, precise, and cost-effective improvement of key economic traits, supports the identification of major genes, and enables the development of targeted breeding strategies, thereby enhancing the genetic potential of the national herd. Moreover, the integration of genomic information into national performance testing programs and the expansion of reference populations can increase selection efficiency and long-term productivity. Thus, breeders should consider the supremacy of GS and adopt an appropriate prediction approach on the basis of their specific objectives. GP methods such as ssGBLUP, WGBLUP, WssGBLIP, and Bayesian have already proved their superiority in terms of carcass and meat quality traits over PBLUP. Despite these rewards, several challenges remain to be overcome. These include maintaining a balance among traits under selection, preserving genetic diversity, and ensuring the long-term sustainability of breeding programs. Addressing these issues requires advanced genomic approaches that consider population structure, gene-environment interactions, and the polygenic nature of complex traits [51]. Consequently, future research should be built on this framework to further exploit the incorporation of high-throughput phenotyping, multi-omics data, and robust statistical models in GP to maximize genetic enhancement while ensuring high-quality beef production. Furthermore, there is much room to investigate ML, such as the application of deep learning to HCs, as a novel idea in the beef industry, which will continue to propel the inventive advancement of Hanwoo breeding schemes.

In conclusion, the genomic evaluation has significantly enhanced the breeding value prediction accuracy for economically important carcass traits. This review confirmed that all the GP models outperform the PBLUP method in the EBVs, and the integration of GBLUP and its extensions within the national evaluation system provides a powerful framework for accelerating genetic gain in the Hanwoo population. These GP approaches allow breeders to make early and more accurate selection decisions, particularly for traits like CWT and MS that are difficult to evaluate on live animals. The polycimaker must give strategic attention to developing the large and well-structured reference population and ensuring the routine gynotyping of breeding animals to sustain the accuracy of GP in Hanwoo. Over the next 5–10 years, the incorporation of whole-genome sequence data, the expansion of genomic evaluations to include reproductive and feed-efficiency traits, and the adaptation of AI-based prediction systems are expected to further strengthen the long-term sustainability and competitiveness of the Hanwoo breeding program.