Skip to content

Multivariate

Reduces the samples-by-features matrix to two dimensions so you can see which samples resemble each other.

PCA scores of the demo samples, with confidence ellipses
PCA scores of the demo samples. Ellipses mark each group's 2σ confidence region.

Controls

Method

  • NMDS (default): non-metric multidimensional scaling on Bray-Curtis dissimilarity, using raw abundances. It is rank-based, so the axes have no percentage; the title reports the stress instead.
  • PCA: principal component analysis on mean-centred, unit-variance-scaled features. Axis labels give the percentage of variance each component explains.
  • PLS-DA: partial least squares discriminant analysis, supervised by biological group. Use it when PCA shows overlapping groups but you expect a real difference and want the direction that separates them best.

View

  • Scores (default): one point per sample, coloured by group, with an ellipse marking each group's 2σ confidence region.
  • Loadings: the 25 features that contribute most to the two displayed components, as arrows from the origin. A feature selected elsewhere is marked here too.
PCA loadings: the features that drive the separation
PCA loadings: the 25 features that contribute most to the two displayed components.
  • Biological Replicates (default): technical replicates are averaged, one point per sample.
  • Individual Injections: one point per injection, useful for spotting a single bad injection.

Clicking a score point selects that injection, which is then shown in the status bar.