EigenSpace

A standalone R/Shiny app that takes expression data and optional metadata, runs PCA, and produces a PowerPoint deck, 600-dpi figures, CSV tables, and a manuscript-ready paragraph.

Structure

EigenSpace/
├── app.R              # Shiny app
├── run_pca.R          # R script to run PCA without GUI/Shiny
├── R/
│   ├── pca_setup.R   # read & align inputs, preprocess, prcomp, result tables, trait tests
│   ├── pca_plots.R    # generate plots using ggplot2
│   ├── pca_summary.R  # manuscript paragraph & slide bullets
│   └── pca_export.R   # build run, save 600 dpi PNGs + CSVs + officer pptx, zip
└── data/
    ├── demo_expression.csv   # 6 genes x 8 samples (example, demo)
    └── demo_metadata.csv     # Group (categorical), Batch (categorical), RIN (quantitative)

The setup, plots, summary, and export functions have no Shiny dependency, so you can source() them and call build_run() / export_run() from any script.

Install

install.packages(c("shiny", "ggplot2", "officer"))

prcomp (the core computation) is part of base R stats; nothing to install.

Run the app

From inside the EigenSpace/ folder:

shiny::runApp()

Click Load demo data to try it immediately, or upload your own files.

Run from a script (no GUI required)

Edit the settings block at the top of run_pca.R, then:

Rscript run_pca.R

Input format

• Expression CSV: first column = IDs. Use the orientation toggle to say whether rows are features or samples. (Demo is features × samples.)
• Metadata CSV (optional): first column = sample IDs matching the expression sample IDs exactly. Remaining columns are traits; the app auto-detects categorical vs quantitative and lets you override.

Outputs (one zip folder)

This part takes a while to show up so you can Save as a zip folder. However, when you save it and unzip the folder, you see the following structure to files and folders:

<project_name>/
├── <project_name>_PCA.pptx   # title, methods (bullets), results, per-trait slides
├── figures/              # every plot as a 600-dpi PNG
├── tables/               # variance, scores, loadings, ANOVA, trait–PC correlations (CSV)
└── summary.txt           # summary/manuscript paragraph

If you would like an overview of PCA based on the demo dataset of 8 samples and 6 genes, see Understanding PCA.

Notes

• Core PCA: stats::prcomp(); Visualization: ggplot2; Deck: officer.
• log2 uses log2(x + 1) so zeros stay finite
• Categorical traits: grouped score plots (with/without 95% ellipses) + per-PC violin/box with one-way ANOVA.
• Quantitative traits: gradient score plots + Pearson trait×PC correlation table and heatmap