Enhancing genomic prediction for key production traits in chickens through ultrasound phenotyping and multi-model comparative analysis

Background Growth performance and carcass traits are economically vital in poultry breeding. In Wenchang chickens, reducing excessive abdominal fat represents a critical breeding objective. However, as a typical carcass trait, abdominal fat thickness has traditionally been measurable only post-slaughter, resulting in inefficient and costly selection processes that hinder genetic progress for these traits. To overcome this limitation, we developed an integrated approach combining non-invasive ultrasound phenotyping and multi-model genomic selection to evaluate growth and fat-related traits in Wenchang chickens. Results We genotyped 3,737 chickens using the “Jingxin No.1” 55K SNP array and performed longitudinal measurement of abdominal fat thickness (AFT) via ultrasound imaging. A comprehensive evaluation of genomic prediction models revealed that WGBLUP (informed by wssGWAS), and GBLUP models based on LD-pruned whole-genome sequencing (WGS) data significantly outperformed standard GBLUP, with accuracy gains of 5.25% and 6.58%–15.30%, respectively. Among the machine learning algorithms tested, kernel ridge regression (KRR) and support vector regression (SVR) achieved the highest predictive improvement (3.00%–4.15%) while maintaining superior computational efficiency, whereas ensemble methods provide no consistent advantage. Conclusions Our work established ultrasound imaging as a scalable, non-invasive phenotyping platform for poultry breeding. Results demonstrated that integrating wssGWAS-derived biological priors with WGS data substantially improves genomic prediction accuracy for complex traits. This integration, enhanced by computationally efficient machine learning algorithms, provides a powerful and practical strategy to accelerate genetic gain.