25 Prediction of Residual Feed Intake in Feedlot Steers Based on Phenotypic Associations with Feeding Behavior and Carcass Ultrasound Traits

• Published in: Journal of animal science 2018-03, Vol.96 (suppl_1), p.13-14
• Main Authors: Parsons, I L, Johnson, J R, Kayser, W C, Miller, M D, Carstens, G E
• Format: Article
• Language: English
• ISSN: 0021-8812; 1525-3163
• DOI: 10.1093/jas/sky027.026

Abstract Previous research has demonstrated that cattle with divergent RFI have distinctive differences in feeding behavior patterns, and carcass fat content. Objectives of this study were to develop and validate predictive equations for RFI utilizing feeding behavior and carcass ultrasound traits as independent variables. Three consecutive-year trials were conducted using Angus crossbred steers (N = 508) with an initial age of 293 ± 18 d and BW of 309 ± 56 kg. For each study, steers were blocked by BW and randomly assigned to 1 of 2 pens equipped with GrowSafe feed bunks. DMI and feeding behavior (FB) traits were measured daily for 70 d while fed a grain-based feedlot diet. Initial and final carcass ultrasound measurements (intra-muscular fat, backfat (BF) depth, rib eye area) were collected on days 0 and 70, and BW was measured at 14-d intervals. Fourteen FB traits were evaluated including frequency and duration of bunk visit (BV) and meal events, head-down duration (HDD), time to bunk, and the corresponding day-to-day variances (SD) of these traits. Additionally, HDD per meal event and BV per meal were included. RFI (0.00 ± 0.78 kg) was calculated within trial by linear regression of DMI (10.1 ± 1.1) on ADG (1.71 ± 0.27 kg/d) and mid-test BW0.75. Separate RFI prediction equations were developed using partial least squares procedures with FB and FB plus ultrasound (FB+) traits included as independent variables. Coefficients of determination for calibration (R2c) were 0.42 and 0.46 for FB and the FB+ models respectively. For the FB+ model, traits with variable of importance (VIP) scores >1 included HDD, BV duration, daily HDD SD, HDD per meal, daily BV duration SD, BV frequency, and gain in BF. Two validation methods (cross and test-set validation) were used to evaluate the accuracies of the RFI predictive equations. The accuracies of cross validation (R2cv) were 0.41 and 0.46 for the FB and FB+ models, respectively. For test-set validation, pen 1 was used for calibration and pen 2 for validation, and the reciprocal. Accuracies for test-set validation (R2v) were 0.32 and 0.39 for FB models and 0.38 and 0.45 for the FB+ models. These results indicate that phenotypic RFI can be predicted based on feeding behavior traits with moderate accuracy. More studies are warranted utilizing larger databases to develop more robust equations for the accurate prediction of individual-animal RFI.