Skip to content

PCA and Exploratory Analysis

Principal Component Analysis reduces a spectral matrix into latent variables that explain major sources of variation.

Use PCA For

  • checking sample clustering
  • spotting outliers
  • assessing preprocessing
  • inspecting loadings
  • finding instrument or batch drift
  • deciding whether calibration or classification is plausible

Outputs to Inspect

  • scores: sample positions in PC space
  • loadings: spectral variables contributing to each PC
  • explained variance / scree
  • Hotelling T2 and Q residual diagnostics when available

Practical Interpretation

Scores show sample relationships. Loadings explain which spectral regions drive those relationships. A good PCA review looks at both, not only the score plot.

In the Workbench, use Color by metadata and Symbol by metadata to switch among available sample-table fields such as specimen, block, operator, batch, or an author-supplied status. The styling changes only the graph; it does not refit PCA or change scores. Missing or misaligned sample identity is refused rather than guessed from filenames or row order.