Deliverable D3.5 A model to predict the variation in nutrient utilisation for different purposes (i.e., growth, gestation, milk and egg production) in monogastric animals

Objectives

The purpose of the model is to quantify the variability in individual trait responses within a livestock population whose individuals share a common factor, e.g., are of the same breed, live in the same farm, or are offered the same feed. In addition to assessing trait variability within a given a population, the approach will also allow comparisons between populations, e.g., that differ in species, breed, environment, or feed, accounting for both their average and their inherent variability in trait responses. The Deliverable has the following objectives towards these aims:

1. To characterise and estimate trait variability from a dataset on a herd, flock or breed.
2. To predict and summarise the influence of variation in model-derived traits on trait variation in a production system.
3. To demonstrate the approach on different monogastric species and on performance and reproduction traits.

Rationale

We developed a modelling approach that has multiple applications in precision livestock farming, nutrition, and selective breeding. The approach is demonstrated through three case studies on monogastric livestock: growth in pigs, and reproduction in sows and in laying hens. Comparison of trait variability across these species and traits shows common aspects that they share as well as their distinctive features. This methodology comprises a data-driven (top-down) approach, where models are fitted to phenotypic trait data obtained from multiple individual animals; and a simulation (bottom-up) approach, where population phenotypic variation is derived and summarised by the average and deviation (i.e., median and confidence interval) for each modelled trait. The approach has the following benefits in relation to current alternatives: 1) Making no prior assumptions about the distributions of traits and their correlations within the population; specifically, it is assumed that the population traits are distributed according to the trait distribution in the group of sampled animals (nonparametric approach). 2) Being computationally faster than current parametric approaches; specifically, the distribution of traits in the wider population is inferred from that in the sample through a process of individual resampling. 3) Having no specific requirements on the size and quality of the datasets input.