Predicting complex traits using microbiome information: A comparison of metagenome distance matrices

Abstract

The aim of this study was to compare ordination methods for microbiota distance matrices, in order to estimate variance components for complex traits prediction including the microbiome in its estimation. Seven ordination methods for building distance (or dissimilarity) matrices were tested; real (n=70) and simulated (n=1000) data were analysed to estimate variance components including phenotypes, genotypes and microbiome information. The seven methods were: the one reported in Ross et al. (2013), multidimensional scaling (MDS), principal coordinates analysis (PCoA), detrended correspondence analysis (DCA), non-metric multidimensional scaling (NMDS), redundancy analysis (RDA) and constrained correspondence analysis (CCA). MDS and PCoA methods yielded exactly the same matrix. From simulation analysis it can be inferred that ordination methods of MDS/PCoA, RDA and CCA are as suitable as or even better than previously reported by Ross et al. (2013) for prediction of microbiability, contrasting with DCA and NMDS which performed poorly as predictive methods for microbiability. From real data analysis, low heritability for feed efficiency and microbiability were obtained and mid to high correlations between the genetic background of the hosts and the phenotypes or microbiota and phenotypes were obtained with methods that performed better in the simulation, indicating that it might be a relationship linking genotype-microbiome-phenotype which could be used in prediction of complex traits.