Multi-omics analysis

Genomic prediction of methane production and dry matter intake using models combining genomic and mid-infrared spectral effects in dairy cattle

---

Abstract

Infrared spectroscopy has been proposed as an alternative to genetic markers for capturing genetic similarities between individuals and improving predictive accuracy. In dairy species, mid-infrared spectra (MIRS) of milk are routinely collected and used to predict complex traits, such as fat and protein yields and methane production. This study integrated a milk MIRS-based similarity matrix, either alone or combined with SNP data, into statistical models for evaluating methane production (MeP) and dry matter intake (DMI). We evaluated (i) the possible association of MIRS with MeP and DMI phenotypes through variance component estimation, and (ii) their impact on genomic estimated breeding value (GEBV) ranking by comparing genomic models with and without MIRS. MeP (g/d) and DMI (kg/d) were recorded over five consecutive days between 110-210 days in milk (DIM) for 639 first-parity Canadian Holsteins cows. A single averaged phenotype record cow was analyzed for each trait. MIRS data were reduced to 512 wavenumbers by removing water absorption noise. The spectral similarity matrix was computed as WW'/n, where W is a matrix of centered and standardized wavenumbers, and n is the number of wavenumbers. Bayesian reproducing kernel Hilbert spaces models were fitted (1,000,000 iterations, burn-in = 100,000, thinning = 10). Three models were compared: genomic-only (G), spectral-only (S), and combined (G+S). Fixed effects included herd, age at calving class, DIM class and year-season of the measurement. Prediction efficiency was evaluated using 5-fold cross-validation. Model performance was compared using the Deviance Information Criterion (DIC) and the predictive ability, defined as the Pearson correlation between genomic- or spectral-predicted values and adjusted phenotypes, within each testing set. Spearman correlations were estimated between GEBVs from G and G+S models. MeP averaged 478.89 ± 80.63 g/d and DMI averaged 19.86 ± 2.22 kg/d. The G+S models showed slightly lower DIC for MeP and similar DIC for DMI, indicating marginal improvement over G models. Heritability estimates for the G model were moderate: 0.38 ± 0.01 for MeP and 0.40 ± 0.01 for DMI. The G+S model explained a greater proportion of the phenotypic variance than G or S models for both traits (0.57 ± 0.02 for MeP and 0.45 ± 0.02 for DMI). For DMI, genomic variance (1.67 ± 0.05) accounted for most of the combined variance (2.24 ± 0.11) in the G+S model, indicating minimal spectral contribution to a trait weakly correlated with milk composition. Rank correlations between GEBVs from the G and G+S models were ≥ 0.94 for both traits, suggesting that including spectral information had small changes in GEBV ranking. The combined models improved predictive ability by +0.02 to +0.12 compared to genomic-only models for both traits. These results warrant further investigation on how spectral data may aid genomic prediction in dairy cattle.

Keywords: 2026

How to Cite:

Machefert, C., Schenkel, F., Baes, C. & Miglior, F., (2026) “Genomic prediction of methane production and dry matter intake using models combining genomic and mid-infrared spectral effects in dairy cattle”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285963. doi: https://doi.org/10.31274/wcgalp.23825

Rights: 1