Recent talks

This work aims to describe existing metabolomics integrative analysis techniques and their importance for animal production. Overall, we aim to demonstrate its relevance in answering complex questions on how animal production could benefit from these techniques. We also present a case study of heat stress in dairy small ruminants that integrates metabolomic data with physiological data. To the best of our knowledge, no reported study in animal production has attempted such integration. We used a univariate mixed model to estimate the marginal posterior distributions of environmental temperature, time of the day, week and period with the productive and physiological variables. Regarding metabolomic data set, the first step carried out was the summary of the data imported into R for quality control. Afterwards, data are exported to MetaboAnalyst and a projection to latent structures-discriminant analysis to extract via linear combination of original variables the information that can predict the class membership. The integration method was an unsupervised multiple kernel learning for heterogeneous data integration. The method is able to combine several kernels into one meta-kernel in an unsupervised framework. On average of BT was greater in HS compared to TN goats. HS depressed DMI, lost BW, decreased milk fat, protein and lactose content compared to TN goats. Heat stress did not affect milk SCC. No differences were observed between TN and HS goats in blood albumin concentration. the HS lowered blood NEFA, and increased BHBA. Furthermore, blood insulin concentration was not affected by HS conditions. Regarding metabolomics data, the cross-validation of plasma metabolomics PLS-DA models (first 2 components) gave Rx2 = 0.80, Ry2 = 0.90, and Q2 = 0.30. Furthermore, identified 15 metabolites with a VIP score > 2.0. The integration of HS and TN datasets shows that the variability between these samples is mostly driven by BT and RR rather than productive, 1HNMR or blood biochemical factors. the most correlated kernels are the blood biochemistry and 1HNMR kernels and that, BT and RR provides a different image of the impact of these variables on productive variables. Mixed models can analyze non-standard models separating clearly the fixed and random effects. In high dimensional data, the classical linear regression is limited, and the multivariate methods are the most appropriate. The big amount of data acquiring technologies makes the proposition of a single data integration methods almost impossible. The data integration methods described in this work are recommended to detected patterns of correlation between data sets of interest for further processing. Data integration methods allows the variables of interest in the huge set of variables measured in an animal production experiment.