Beta diversity (or Ordination) analysis

Beta diversity characterises differences in the ARG or resistome composition between samples. There are several beta diversity indices/measures that can be used to generate a distance or dissimilarity scores to compare the feature dissimilarity between pairs of samples. The resulting scores can be combined into a distance matrix and used for ordination to visualise patterns in our data. Samples close to each other are more similar to each other in their resistome composition.

Currently, the five most commonly used beta diversity measures are supported in ResistoXplorer, with each having its own way of calculating dissimilarity between samples as they assess different aspect of resistome composition. For example, ‘Bray-Curtis dissimilarity’ uses abundances information of features (ARGs) to calculate differences in resistome composition. While, ‘Jaccard distance’ only uses just the presence or absence (occurrence) of features for estimating dissimilarities. Similarly, ‘Jensen-Shannon divergence (JSD)’ calculates the distance between two probability distributions that account for the occurrence of features in data.

Beta-diversity measures can be visualised using various ordination methods. Ordination are the techniques to summarise and project multidimensional data into lower-dimension (2-3d) space. Currently, three widely accepted ordination methods are supported in ResistoXplorer to explore and visualise differences between resistome, including principal coordinate analysis (PCoA), non-metric multidimensional scaling (NMDS) and principal component analysis (PCA). PCoA and NMDS methods take the distance matrix as an input and are sensitive to distance method used. The PCoA maximises the linear correlation between samples, wherein NMDS maximises the rank-order correlation between samples. Additionally, in case of NMDS, data is not required to fit a normal distribution. One should choose PCoA if distances between samples are so close that a linear transformation would be enough. While, NMDS is recommended to highlight the gradient structure within the data. Another common ordination method is PCA, which is just a type of PCoA that uses Euclidean distance. Using this method with centered log-ratio transformed data it is an alternate ordination approach (CoDA-based) that can able to handle the issue of compositionality.

Additionally, we can also assess the statistical significance of beta diversity clustering between the groups using either PERMANOVA (PERmutation Multivariate ANalyses Of Variance), analysis of group similarities (ANOSIM) or homogeneity of group dispersions (PERMDISP). These tests evaluate global differences in resistome composition between two or more groups. PERMANOVA tests whether the centroids of all groups are equivalent. It uses the distances (or dissimilarity) between samples of the same group and compares them to the distances between groups. Though, this method is sensitive to multivariate dispersions. Consequently, PERMDISP should also be used to test whether the dispersion (or variation) between samples differs from the dispersion between groups. While, ANOSIM evaluates whether within-group distances or dissimilarities are greater or equal to between-group distances using the ranks of all pair-wise sample distances.

Similar to alpha diversity, ordination analysis can also be performed at different functional levels in ResistoXplorer. This analysis is performed on processed (filtered and normalised) data.

Now let’s go to the next step in order to learn how to do such analysis in ResistoXplorer.