Leveraging ultrasonic-derived phenotypes and estimated breeding value to improve abdominal fat weight prediction in chickens throughout the egg laying period

Abdominal fat (AF) in hens impacts egg production and may reflect poor feed efficiency, meaning that dynamic monitor of AF changes facilitate to optimize feeding management and production efficiency. This study estimated hens’ AF weight among whole laying period through fitting vivo phenotypes by te...

Full description

Saved in:
Bibliographic Details
Main Authors: Penghao Li, Zhengda Li, Fan Ying, Dan Zhu, Dawei Liu, Xianyi Song, Jie Wen, Guiping Zhao, Bingxing An
Format: Article
Language:English
Published: Elsevier 2025-08-01
Series:Smart Agricultural Technology
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2772375525001455
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abdominal fat (AF) in hens impacts egg production and may reflect poor feed efficiency, meaning that dynamic monitor of AF changes facilitate to optimize feeding management and production efficiency. This study estimated hens’ AF weight among whole laying period through fitting vivo phenotypes by ten machine learning techniques (including generalized linear model, GLM; multiple linear regression, MLR; ridge regression, RR; LASSO; elastic net, EN; k nearest neighbours, KNN; SVM biased linear and Gaussian kernel; Random forests, RF and XGBoost). Consistent protocols were applied to all phenotypic measurements to minimize batch effects across five laying stages (8 traits: abdominal fat thickness (AFT), keel length (KL), breast width (BW), pubic bone width (PBW), body slope length (BSL), live weight (LW), keel-pubic length (KPL), and abdominal fat weight (AFW). The stepwise backwards variables selection was conducted to rule out the possible bias of multicollinearity between independent variables and AFW. While, AFT measured by ultrasound improved the predictive ability of all the models (R² of KNN showed highest increase of 12.35 %). Moreover, we also evaluated the contribution of AF estimate breeding values (EBVs) to fitting model performance. When incorporating EBVs of AFW as extra independent variables, all methods’ predictive ability had increased by an average of 15.71 %, especially KNN (27.2 %). For muliple laying periods, the EN model showed the best performance at 26 and 35 weeks, with an average R² of 0.928, MAE of 6.709, RMSE of 9.862, and MAPE of 9.660. The KNN model performed best at 28, 31, and 43 weeks, with an average R² of 0.918, MAE of 9.273, RMSE of 12.348, and MAPE of 17.240. These findings underscore the feasibility of accurately prediction of hens'AF through fitting appropriate algorithms for different laying periods, that supporting delicacy feeding management in farm.
ISSN:2772-3755