PCA GUI

Desktop GUI (Python/tkinter) for Principal Component Analysis on tabular data — e.g. compound/ metabolite concentration profiles across sample groups (HPLC data, chemotaxonomy, quality control), or any samples × variables dataset — producing publication-ready plots: scree plot, loadings bar chart and biplot.

Features

• Load data from Excel, in either orientation — samples as rows or samples as columns (with automatic transposition, see trasponi.py)
• Choose which variables to include and which samples to exclude, with per-group colors and distinct markers (up to 6 groups)
• Several scaling methods: z-score, mean-centering, min-max, robust, or none
• Automatic handling of missing values (mean imputation, with a warning)
• Three linked plots in one figure: scree plot (explained/cumulative variance), loadings bar chart, and biplot (scores + loading vectors + group confidence ellipses)
• Editable spreadsheet view of the loaded/transformed data
• Export: full figure or individual panels (PDF/PNG/SVG), scores/loadings/variance to Excel, or a ready-to-plot Excel workbook (separate sheets for scores, loading arrows, group ellipses) for building the same chart natively in Excel
• Edit Figure... reopens any one panel (scree / loadings / biplot) in PlotStyleKit's standalone figure editor for titles, per-element styling and publication-size export; Salva figura (pickle)... saves a panel as a live, re-editable Figure object instead of a raster image — see Requirements below

Requirements

pip install -r requirements.txt

numpy, pandas, scikit-learn, matplotlib, openpyxl
adjustText   # optional — avoids overlapping labels in the biplot

PlotStyleKit is an optional sibling repo (cloned as ../PlotStyleKit next to this project) providing the shared Origin-like plot style and the standalone figure editor behind Edit Figure... / Salva figura (pickle)...; without it those two buttons are unavailable and the app uses default matplotlib styling.

See installa_pca_gui.txt for a step-by-step Windows/PowerShell setup guide (Italian).

Running

pythonw pca_gui.pyw

Double-clicking the .pyw file on Windows works too (no console window).

Input file formats

Standard layout — first row is the header, one sample per row:

Sample | Group | Variable1 | Variable2 | …
F1     | F     | 38.5      | 466.4     | …
F2     | F     | 0.03      | 370.7     | …

Transposed layout — samples arranged in columns, detected and converted automatically:

Samples  | F1   | F2   | … | SD6
Groups   | F    | F    | … | SD
Compound1| 38.5 | 0.03 | … | 0.07
Compound2| 466  | 370  | … | 240

trasponi.py converts a transposed file to the standard layout on its own (useful for sharing data with colleagues):

python trasponi.py <transposed_file.xlsx> [output_file.xlsx]

Tests

python -m unittest discover -s tests -v

The tests open a hidden Tk window (they are skipped when tkinter or a display is not available), run the PCA on the built-in example data and check the results against scikit-learn: explained variance, orthonormal loadings, group separation, excluded samples, missing-value imputation and the scaling options. They also check that the standard and transposed layouts give the same table and that the app runs when the PlotStyleKit sibling repo is missing.

Security note

Salva figura (pickle)... writes Python pickle files, and Edit Figure... in PlotStyleKit opens them. A pickle can run arbitrary code when it is loaded, so open only figure files you created yourself or received from someone you trust.

Structure

• pca_gui.pyw — main application (a single PCAApp class)
• trasponi.py — standalone converter, transposed layout → standard layout
• installa_pca_gui.txt — Windows setup guide (Italian)
• tests/test_pca_core.py — unit tests (see Tests)
• requirements.txt — Python dependencies

License

MIT