Spikeless

Desktop GUI to clean single-point spikes from radio-HPLC spectra exported by GINA X, produce publication-ready plots, and analyse/export them. MIT-licensed (see LICENSE).

Run

Double-click Spikeless.bat (the launcher). On first run it downloads uv, installs a private Python + dependencies under C:\Users\Public\Spikeless (no admin rights, no PATH changes), drops a Spikeless shortcut (with the app icon) on your Desktop, then opens the GUI with no console window. Later runs start immediately — use the Desktop shortcut.

To remove everything it installed, run uninstall.bat (deletes that one folder; your project files and exports are untouched — Spikeless makes no PATH or registry changes).

Use

1. Drag a GINA .txt onto the window (or Browse…). The spectrum plots immediately. Before any data is loaded there is no plot, and Plot options / Detect are disabled.
2. Data panel is a tree. Each dataset node is metadata only — name, file, run date/time, molecule, radioisotope (dropdown), and user-added conditions (Info, each a free note or a label/value pair). Under it sit curves: the original, and processed curves you create (spikeless, decay-corrected, …). Under each curve sit its results — a spikes group (→ each spike, showing y), a baseline, and a peaks group (→ each peak with Rt / y / AUC / %). Every node has a checkbox (unticking a parent unticks its children), can be renamed (double-click — a pen marks editable rows) and removed (✖). Drag curves/datasets to reorder them = change draw order (top of tree draws under, so later items sit on top — matters with alpha). Drop a dataset onto another dataset, or a curve onto a sibling curve, to reorder; a drop line shows where it will land. Buttons act on the selection: Info (details for any element — dataset metadata, curve, spikes, baseline, peaks, or a single spike/peak), Adjust (normalization, curve only), Display… (any node — appearance), Duplicate (curve).
3. Processing acts on the selected curve (so detecting on a spikeless curve finds no spikes). Pipeline order is spikes → decay → baseline → peaks: spikes are an electronic artefact unrelated to sample activity, so they are removed before decay correction; decay comes before baseline/peaks so AUC/% reflect the activity-corrected signal. Each action has a ⚙ for its parameters:
   • Detect spikes → marks spikes and adds a spikeless child curve (⚙: window, threshold, max width, interpolation linear / PCHIP).
   • Apply decay → adds a decay-corrected child curve. Its ⚙ sets the isotope (+ half-life) and the time to decay-correct to (default run start); if the isotope/half-life is missing when you click Apply decay, that dialog opens automatically. (Hidden by default — enable it under Options → Processing features shown.)
   • Detect baseline → adds a baseline (⚙: minimum / average of first N).
   • Detect peaks → adds peaks with Rt/AUC/% (⚙: prominence, height, distance, and the local drift baseline used for integration). Detection runs on a spike-suppressed copy of the curve, so spikes don't become skinny false peaks. Running it on a plain curve makes a fresh peaks group; running it on an existing peak/peaks group reprocesses with current options.
4. Plot options — a collapsible panel down the left side (closed by default; open it with its toolbar button). Sections: Plot area, Margins, Legend, Y axis, X axis — plot size in mm, plot-area/margin background colour + alpha, legend position/font, axis limits, fonts, tick length/thickness, and each axis's Advanced (title/label spacing, and a grid on major + minor). (Per-curve appearance is set from the Data tree via Display….) Numeric axis limits, graduation start/end, and tick thickness have Auto tickboxes (auto tick thickness = axis thickness); a ticked Auto limit shows the value auto-mode would use, greyed out. Each axis has an Advanced sub-section (title↔label and label↔axis spacing, and a grid). All sections start collapsed except Plot area.
5. Copy / Export / ⚙ live per window, not on a shared toolbar:
   • Log and Report carry Copy · Export · ⚙ · ✕ on their title bar. The ⚙ holds that export's own options (log: include timestamps; report: copy tables as rich HTML). The report additionally shows a small ⧉ copy table link next to each peak table in the text — click it to copy that one table (there can be several).
   • The plot has a ⧉ Copy · ⭳ Export · ⚙ cluster in the top-right of the graph area; its ⚙ sets PNG file dpi, clipboard dpi (default 600), and clipboard format (PNG / SVG). Export writes PNG (default 600 dpi) or SVG at the exact mm size, transparency preserved. The report ⚙ also has Export selected curve (GINA X) — writes the processed signal back to a .txt in GINA format so it can be reloaded (best-effort GINA X compat).
6. Options — collapsible sections (all start collapsed): app background (checker/solid, cell size in plot mm), a dotted export-area border, Plot view (resolution-slider max), Processing features shown (tick which Processing buttons appear — Apply decay off by default), and Menu windows (which docks open by default, plus lock menu windows). These preferences persist between sessions.

The Plot options / Data / Processing / Log / Report toolbar buttons open and close their docks; closing a dock un-presses its button, pressing it again reopens it.

Plot navigation

The plot is rendered as a vector image you can zoom and pan: mouse wheel zooms centred on the cursor, drag (any button) pans, double-click fits to the window. Drop a GINA .txt straight onto it to load. The resolution slider (bottom-left) enlarges the plot area 1×–5× on screen (fonts/line widths stay fixed, so closely-spaced spikes/peaks separate out) — tick export to apply it to exports too. The ⧉ Copy button (top-right cluster) copies the plot to the clipboard at its real mm size (the image carries its dpi, so pasting into PowerPoint gives the Plot-options size, not a giant); clipboard dpi (default 600) is set in the plot ⚙.

Adjust (decay correction & normalization)

Per-dataset and fully reversible (it never mutates the loaded data):

• Decay correction — corrects each sample back to a reference time using the isotope half-life. The half-life is auto-filled from a built-in table for the Radioisotope field (Lu-177, Ga-68, F-18, …) and can be overridden manually. Reference = run start (from the file header), an offset in minutes from start, or a clock time on the run date.
• Normalization — Counts (cps), % of max, or % of total. The % modes subtract a baseline first; baseline method is selectable (Minimum, Average of first N points; the estimator is a small registry so drift-aware / segmented methods can be added later).

Spike detection algorithm

A spike is a single acquisition point whose value sits far above the local trend. Detection is a Hampel filter (rolling median + median absolute deviation), which is local, so it flags a spike even when it rides on a genuine chromatographic peak while leaving the multi-point peak itself untouched. For signal y and window w (odd; default 7):

1. Local reference med = median_filter(y, w) (edges reflected).
2. Local scale from the residual: mad_local = median_filter(|y − med|, w), converted to a robust standard deviation σ = 1.4826 · mad_local (1.4826 makes MAD ≈ σ for Gaussian noise).
3. Two noise floors keep quiet regions from over-flagging:
   • global floor: σ = max(σ, 1.4826 · median(|y − med|)), so a locally flat window whose mad_local → 0 doesn't flag ~1-count wiggles;
   • Poisson floor (count data): σ = max(σ, sqrt(max(med, 1))), the shot-noise scale of cps.
4. Threshold: a point is a candidate when y − med > n_sigma · σ (n_sigma default 5; positive_only flags only upward excursions — real spikes are high cps).
5. Width guard: any run of consecutive candidates longer than max_width (default 1) is dropped, because a wide excursion is a real peak, not a spike.

Parameters (⚙ next to Detect): window, n_sigma, max_width.

Spike removal (interpolation)

Flagged points are replaced by interpolation from their nearest un-flagged neighbours; all other samples are left exactly as-is, so a point on a real peak's flank is restored to the local peak level, not to baseline. If fewer than two clean points remain, the signal is left unchanged. Method (Remove ⚙):

• Linear (default) — a straight line between the nearest clean points (numpy.interp). For a single-point spike that is exactly n-1 → n+1. Cannot overshoot.
• PCHIP — monotone cubic interpolation across the gap. Smoother for multi-point gaps and, unlike a plain cubic spline, will not overshoot into ringing on a peak flank.

Detecting spikes never mutates the loaded data: it marks a spikes group on the curve and adds a spikeless child curve, which you can further process (baseline, peaks, decay) like any other curve. Show/hide, rename, or remove each node in the Data tree.

Peaks (Rt, AUC, %)

Detect peaks on a curve finds local maxima (prominence-based) and computes, per peak, Rt (apex x), x start/end/width, y max (above baseline), AUC, % of the group total, and a skewness. Select a peak → Info shows all of it (name editable); Display… sets the peaks group's style. Integration bounds run apex-to-valley (the low point between a peak and its neighbour) then trim to the peak's feet, so the shaded region spans the whole peak rather than a narrow tip. Peaks draw as a shaded AUC fill by default (or markers), with optional on-graph Rt / % labels. The baseline used for integration is a local drift line between each peak's bounds by default, or the curve's detected baseline — a Detect peaks ⚙ option. The report is the dataset metadata + conditions + a per-curve peak table; a ⧉ copy table link next to each table copies an HTML table you can paste straight into Word/PowerPoint.

GINA format notes

• Windows-1252 encoded, tab-separated, decimal comma.
• Signal = 2nd column (radiodetector count rate, cps).
• The time column is display-formatted and lossy, so time is reconstructed assuming uniform sampling (interval inferred from the file; all known files are 1 Hz).

Dev

uv run python -m spikeless              # launch
uv run python -m spikeless.spikes       # spike-detection self-check
uv run python -m spikeless.peaks        # peak-detection self-check
uv run python -m spikeless.adjust       # decay/baseline/normalization self-check
uv run python -m spikeless.io_gina      # parser + GINA-save round-trip self-check
uv run python -m spikeless.plotting     # plot-sizing self-check

Roadmap (phase B)

More baseline estimators — segmented and asymmetric least squares (arPLS). Richer peak parameters.

Done: SNIP drift-aware baseline; per-peak report table (area/height/retention) + Excel export; per-peak colour and on-graph labels; configurable per-button default display actions.