Predicting beef diet nutritional composition and intake from rumen metagenomic profiles

Area of Science:

• Animal Science
• Microbiology
• Bioinformatics

Background:

• Accurate knowledge of beef cattle diet composition and intake is crucial for feed traceability and modeling methane emissions and performance.
• Direct measurement of diet and intake is often costly, labor-intensive, and not always feasible.

Purpose of the Study:

• To investigate the use of rumen metagenomic data and machine learning to predict diet type, nutritional composition, and intake levels in beef cattle.
• To assess the generalizability of these predictive models through external validation.

Main Methods:

• Rumen metagenomic sequencing was performed on samples from 142 beef cattle (Luing and Charolais crossbred) fed different forage-to-concentrate ratios.
• Machine learning algorithms, including random forest, were employed to analyze microbial gene and genus abundances for prediction.
• Models were developed to predict diet type, nutritional components (starch, crude protein, fiber, energy), and dry matter intake (DMI).

Main Results:

• The log-ratio of Verrucomicrobia and Chlorobi abundances accurately discriminated diet type (0.86 ± 0.05).
• Predicting nutritional diet components achieved external validation accuracy between 0.77 and 0.83.
• Microbiome data were more effective for predicting feed composition than DMI (0.27 ± 0.12), though classification into low/medium/high DMI was accurate (0.74).
• Incorporating dietary information improved phenotypic models for methane production (MP) and DMI, and genetic models for MP.

Conclusions:

• Rumen microbiome data, analyzed with machine learning, serve as a valuable tool for post hoc prediction of feed composition in beef cattle.
• This approach can enhance feed traceability and contribute to more accurate modeling of diet impacts on animal performance and environmental emissions.