Genomic prediction integrating indirect breeding values from MIR spectra data in dairy cows
Résumé
Mid-infrared (MIR) spectroscopy provides a high-throughput, cost-effective phenotyping method that captures molecular information from milk samples. This proof-of-concept study evaluates the integration of MIR spectral data into genomic prediction models to improve the estimation of breeding values for production, health, fertility, and morphological traits in three French dairy breeds: Holstein, Normande, and Montbéliarde. A total of 6,861 cows were analyzed, with genotypes from a 50K SNP chip and MIR spectral records. Heritabilities of MIR spectra varied across breeds, with estimates reaching 0.45 in Holstein, 0.42 in Normande, and 0.30 in Montbéliarde. We compared four genomic prediction models: (1) GBLUP (genomic relationship matrix), (2) FBLUP (MIR-based relationship matrix), (3) F+GBLUP (genomic and MIR-based relationship matrices), and (4) GOBLUP (genomic model incorporating indirect breeding values derived from the genetic effects of MIR). Prediction accuracies improved when MIR data were integrated. For instance, in Holstein, protein content prediction accuracy increased from 0.56 (GBLUP) to 0.60 (GOBLUP), while in Montbéliarde, fertility traits showed a gain of up to 0.05 in prediction accuracy. Across-breed prediction was more effective with MIR data than with genomic data alone. These results highlight the potential of MIR spectra as an auxiliary tool to enhance genomic predictions in dairy cattle breeding programs.