Using machine learning to predict feed intakes of meat sheep from animal traits and ruminal microbiota
Résumé
Animal traits, such as body weights, and rumen microbiota composition have been proposed as feed intake predictors. The present study assessed what are the best predictors out of animal traits, metabarcoding data or a combination of both. Predictions were carried out with sparse Partial Least Squares Regression (sPLSR), Support Vector Regression (SVR) and Random Forest Regression (RFR). With all three approaches, best feed intake predictions were obtained with animal traits only. The generalizability of models to animals of an independent year was assessed: negative (<-0.1) to high (>0.8) correlations between actual and predicted feed intakes were obtained. Finally, estimated breeding values (EBVs) were computed for actual and predicted feed intakes. These EBVs were highly correlated (>0.9) depending on the prediction approach. It mainly varied with proportions of true and predicted feed intakes used during the genetic evaluation.
Domaines
Sciences du Vivant [q-bio]Origine | Fichiers produits par l'(les) auteur(s) |
---|