Phenomics

Assessing the genetic basis of digital phenotype for genetic selection in beef cattle

---

Abstract

Phenotypes are the bottleneck in the post-genomic era for accelerating genetic progress in beef cattle, yet we still use pre-genomic methods to measure them. For instance, body weight requires repeated animal handling, specialized facilities, and substantial labor investment. In this study, we investigated the use of imaging technology to generate digital body weight and compare the genetic basis between digital and conventional phenotypes. Yearling body weight was collected from 512 commercial Angus steers. Cattle were video-recorded using a RealSense D455 depth camera mounted over the scale exit. Images were extracted from the videos and processed to measure body-projected volume, surface area, height, length, and width biometric features using an in-house Python code. These biometric traits were used as predictors of yearling body weight through several analytical approaches, including linear regression, partial least squares regression, elastic net regression, random forests, support vector machines, gradient boosting machines, and neural networks. Model performances were evaluated using leave-one-out cross-validation. The coefficient of determination (R2), regression slope, mean absolute error (MAE), and root mean square error (RMSE) were calculated to quantify the agreement between predicted and observed values. Linear regression, partial least squares, and elastic net models showed the same accuracy (R2 of 0.96), the lowest MAE (10.12, 9.98, and 10.06, respectively), the lowest RMSE (12.92, 12.87, and 12.88, respectively), and no bias (regression slope equal to 1). Although the results are similar among these approaches, the linear regression model is simpler to implement, requires no hyperparameter tuning, is computationally efficient, and has direct interpretation, compared to the other models tested. Then, conventional and digital traits (predicted using the linear regression model) were analyzed using a two-trait animal mixed model to assess heritability estimates, the genetic correlation between them, and the estimated breeding values (EBV) for each trait. In addition, the heritability of the measurement error (ME), defined as the difference between the scale-measured and digital body weight, was estimated using a single-trait animal mixed model. Contemporary groups and age at measurement were included as fixed effects in the models. The analysis was performed using the BLUPF90 family of programs. The concordance correlation coefficient (CCC) between the EBV rankings of the digital and conventional traits was also calculated. The heritability estimates were similar for scale-measure (0.47) and digital (0.49) traits. The genetic correlation between the two traits was high (0.99), suggesting they capture the same underlying genetic signal. EBV rankings showed a high CCC (0.98). The heritability of the ME was low (0.001), indicating that the error in predicting the digital trait is non-genetic. Although the sample size is small, these findings demonstrate that digital body weight is genetically consistent with scale-measured body weight and can be used for genetic selection without compromising selection accuracy.

Keywords: 2026

How to Cite:

Abdullah, H., Hanson, E. & Bresolin, T., (2026) “Assessing the genetic basis of digital phenotype for genetic selection in beef cattle”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2287237. doi: https://doi.org/10.31274/wcgalp.24244

Rights: 1