Phenomics

Integrating milk mid-infrared spectral and genomic information to predict methane emission in dairy cows

---

Abstract

Accurate prediction of individual methane (CH₄) emissions under commercial farm conditions is essential for mitigating the environmental impact of dairy production. Such information enables the genetic selection of animals with greater metabolic efficiency and lower environmental impact. It also supports optimized diet formulation and allows large-scale evaluation of cow efficiency in commercial herds. This preliminary proof-of-concept study aimed to develop and compare modeling strategies to predict two methane-related traits: daily methane emission (ME, expressed in g/day) and methane conversion efficiency (YM, percentage of gross energy intake), using milk mid-infrared spectroscopy (MIRS), genomic SNP data, and their integration. Both traits were jointly analyzed due to biological complementarity and strong positive correlation (r = 0.80). The dataset comprised Holstein cows from the National Institute for Agricultural Research (INIA, Uruguay), with methane phenotypes obtained using the sulfur hexafluoride (SF₆) tracer technique. Measurements were collected per animal across three experimental groups (36 animals), reflecting a limited sample size suitable for methodological evaluation. Genomic information was obtained from medium-density SNP panels, and milk spectra were measured by Fourier-transform mid-infrared (FT-MIR) spectroscopy (900-5000 cm⁻¹; MilkoScan, Foss Electric (FOSS)). Spectra were preprocessed using Savitzky-Golay first derivative and standard normal variate transformation. Informative spectral regions were identified using Elastic Net regularization. Predictive models were calibrated using 5-fold cross-validation grouped by animal to avoid within-animal data leakage. Three prediction strategies were compared: (1) MIRS-only (Partial Least Squares, BayesB/C, Random Forest, and Elastic Net), (2) SNP-only (GBLUP), and (3) combined SNP + MIRS (Bayesian multi-kernel and integrated Elastic Net). Predictive ability was assessed using the squared predictive correlation (r²), root mean squared error (RMSE), and mean absolute error (MAE), with phenotypes adjusted for sampling date, milk collection period, and diet. Predictions based on MIRS achieved the highest predictive accuracy using PLS for both traits (ME: r² = 0.864, RMSE = 29.9, MAE = 23.1; YM: r² = 0.887, RMSE = 0.73, MAE = 0.59), outperforming other MIRS-based models (r² between 0.30 and 0.60). Genomic prediction resulted in r² = 0.19. Models integrating SNP and MIRS information via Elastic Net improved prediction (r² ≈ 0.66 for both traits), exceeding SNP-only predictions and approaching MIRS-only performance using the same methodology (r² ≈ 0.60). Bayesian multi-kernel variance decomposition indicated that both genomic and spectral components contributed to the explained variability. Overall, despite the limited sample size, the results indicate that spectrometry data can predict phenotypes related to methane emission. Integrating genomic and MIRS information enhances predictive ability for methane-related traits and provides a foundation for future genetic and nutritional strategies aimed at reducing methane emissions and improving feed energy efficiency in dairy production systems.

Keywords: 2026

How to Cite:

Ferraz, N., Rosa, G., Cartaya, A., Cabrera, J., Ciganda, V., Silva, G., Carracelas, B., Ravagnolo, O. & Aguilar, I., (2026) “Integrating milk mid-infrared spectral and genomic information to predict methane emission in dairy cows”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286318. doi: https://doi.org/10.31274/wcgalp.23923

Rights: 1