MultiGaussFit

About

MultiGaussFit is a desktop GUI application for interactive signal preprocessing and multi-Gaussian deconvolution. Built with Python and Tkinter, it is designed for spectroscopists, analytical chemists, and researchers who need to decompose complex overlapping peaks into individual Gaussian components. It provides a complete workflow from raw data loading through noise filtering, baseline correction, peak fitting, and batch export of results.

---

Key Features

• CSV Data Loading: Supports both XYYY and XYXY column structures with automatic header detection.
• Interactive Plot: Real-time visualization with zoom, pan, and click-to-select functionality powered by Matplotlib.
• Signal Preprocessing:
  • Linear baseline correction
  • Minimum shift (translate signal so min = 0)
  • Negative value removal
  • Savitzky-Golay smoothing filter (configurable window and polynomial order)
  • Fourier Transform noise filter (configurable number of frequencies)
  • Full Undo and Reset support for all preprocessing steps
• Multi-Gaussian Deconvolution:
  • Click on the plot to define deconvolution intervals
  • Fit 1 or 2 Gaussians per peak center
  • Optimization via Differential Evolution (global optimizer)
  • Displays individual Gaussian components, total fit, and SSE quality metric
• Batch Processing: Deconvolve the first signal manually, then automatically apply the same preprocessing pipeline and interval configuration to all remaining signals in the file.
• Real-Time Results: Peak parameters (Amplitude, Center, Width) are displayed live during batch processing.
• Export to CSV: Save preprocessed signals and deconvolution results with full peak parameters.
• Dark/Light Theme: Toggle between dark and light modes for comfortable viewing.

---

Demo

20260506_020400.mp4

---

Workflow

The typical workflow in MultiGaussFit follows these steps:

1. Open a CSV file

Load a CSV file containing your spectral or signal data. MultiGaussFit supports two column structures:

XYYY Structure

A single shared X column (e.g. wavenumber, wavelength, time) followed by multiple Y columns, one per signal:

Wavenumber (cm⁻¹)	Signal_A	Signal_B	Signal_C
4000	0.0236	0.0195	0.0253
3998	0.0238	0.0171	0.0252
3996	0.0224	0.0180	0.0238
...	...	...	...

XYXY Structure

Alternating X–Y column pairs, where each signal has its own independent X axis:

X_Signal_A	Signal_A	X_Signal_B	Signal_B
4000	0.0236	4000	0.0195
3998	0.0238	3998	0.0171
3996	0.0224	3996	0.0180
...	...	...	...

2. Load Data

Select the correct data structure (XYYY or XYXY) and click Load Data. The available signals will appear in the sidebar.

3. Select a Signal

Click on any signal from the sidebar list to visualize it on the main plot.

4. Preprocess the Signal

Apply one or more preprocessing steps as needed:

• Baseline Correction: removes a linear baseline estimated from the signal edges.
• Savitzky-Golay Filter: smooths the curve to reduce high-frequency noise.
• Fourier Transform Filter: filters the signal in the frequency domain, keeping only the dominant frequency components.
• Shift Minimum: translates the signal so that its minimum value equals zero.
• Remove Negatives: sets any remaining negative values to zero.

Use Undo to step back one operation or Reset All to restore the original signal.

5. Define Deconvolution Intervals

Click on the plot to mark the boundaries (start and end) of each peak region you want to deconvolve.

6. Run Deconvolution

Choose 1 or 2 Gaussians per peak and click Run Deconvolution.

Tip: If your peaks are sharp or pointed (i.e. they end in a narrow tip rather than a smooth rounded top), try using 2 Gaussians per peak. The two-Gaussian mode can better capture the cusp-like shape of sharp signals that a single Gaussian cannot reproduce.

7. Inspect Results

The fitted Gaussian components, total fit curve, and SSE quality metric are shown on the plot. Peak parameters (Amplitude, Center, Width) appear in the results panel.

8. Batch Processing (optional)

Click Deconvolve All (same config) to automatically apply the same preprocessing pipeline and deconvolution intervals to every signal in the file. Results update in real time.

9. Save Results

Click Save Results... to export preprocessed signals and deconvolution parameters to CSV files.

---

Output Files

File	Description
{name}_preprocessed.csv	Preprocessed signal values (Position + preprocessed Y for each signal)
{name}_deconv.csv	Deconvolution parameters per peak: Amplitude, Center, Width, SSE

Deconvolution CSV Columns

Column	Description
Signal	Signal name from the original CSV
Peak	Peak number (1, 2, …)
Amplitude (A)	Total amplitude of the peak (in 2-Gaussian mode: $A_1 + A_2$)
Center (µ)	Peak center position
Width (σ)	Peak width (standard deviation)
SSE	Sum of Squared Errors for the fit
a1, a2, sigma1, sigma2	Individual Gaussian parameters (only in 2-Gaussian mode)

---

Installation

Windows Installer (Recommended)

⬇ Download MultiGaussFit_Setup.exe

The installer will:

• Install MultiGaussFit to Program Files
• Create a Start Menu shortcut
• Optionally create a Desktop shortcut

No Python installation required. Everything is bundled in the installer.

Run from Source

Using Conda

git clone https://github.com/1JELC1/MultiGaussFit.git
cd MultiGaussFit
conda env create -f environment.yml
conda activate multigaussfit
python MultiGaussFit.py

Using pip

git clone https://github.com/1JELC1/MultiGaussFit.git
cd MultiGaussFit
pip install -r requirements.txt
python MultiGaussFit.py

Requirements: Python 3.9+

Building from Source

Portable Executable (single file)

pyinstaller MultiGaussFit_onefile.spec

Output: dist/MultiGaussFit.exe (self-contained, ~120 MB)

Fast-Startup Executable (directory)

pyinstaller MultiGaussFit_onedir.spec

Output: dist/MultiGaussFit/ directory with fast startup time.

Windows Installer

Requires Inno Setup 6:

1. Build the onedir version first.
2. Compile the installer script:

   iscc installer/MultiGaussFit.iss

3. Output: dist/MultiGaussFit_Setup.exe

---

Example Data

A sample CSV file is included in the examples/ folder:

python MultiGaussFit.py
# Then open: examples/sample_signal.csv

---

Technical Details

Optimization Algorithm

MultiGaussFit uses Differential Evolution (scipy.optimize.differential_evolution) for global optimization of Gaussian parameters. Differential Evolution is a stochastic, population-based optimizer that explores the entire parameter space without requiring an initial guess. Unlike gradient-based methods (e.g. Levenberg-Marquardt), it does not get trapped in local minima, which makes it particularly robust when fitting overlapping or poorly resolved peaks.

Objective function: The optimizer minimizes the Sum of Squared Errors (SSE) between the observed signal and the sum of Gaussian components within each user-defined interval:

$$\text{SSE} = \sum_{i} \left[ y_i - \sum_{k} G_k(x_i) \right]^2$$

A lower SSE indicates a better fit.

Gaussian Model

1-Gaussian Mode

Each peak is modeled as a single Gaussian function:

$$G(x) = A \cdot \exp\left(-\frac{(x - \mu)^2}{2\sigma^2}\right)$$

Where:

• $A$ = amplitude (peak height)
• $\mu$ = center position
• $\sigma$ = standard deviation (controls the width)

This mode works well for peaks with a smooth, rounded, symmetric shape.

2-Gaussian Mode (for sharp / pointed peaks)

When a peak is sharp or pointed (i.e. it has a narrow, cusp-like tip rather than a smooth rounded maximum), a single Gaussian cannot accurately reproduce its shape. In this case, MultiGaussFit uses two Gaussians sharing the same center position $\mu$ but with independent amplitudes and widths:

$$G_{\text{total}}(x) = A_1 \cdot \exp\left(-\frac{(x - \mu)^2}{2\sigma_1^2}\right) + A_2 \cdot \exp\left(-\frac{(x - \mu)^2}{2\sigma_2^2}\right)$$

Where:

• $\mu$ = shared center position (both Gaussians are centered at the same point)
• $A_1$, $A_2$ = individual amplitudes
• $\sigma_1$, $\sigma_2$ = individual widths (typically one narrow, one broad)

The total peak height at the center is the sum $A_1 + A_2$, which can be used directly for quantification. One Gaussian captures the sharp, narrow core of the peak while the other captures the broader base, together reproducing the pointed shape that a single Gaussian cannot fit.

Preprocessing Pipeline

Step	Equation	Method
Baseline Correction	$y' = y - L(x)$	Estimates a linear baseline $L(x)$ by interpolating between the minima of the left and right halves of the signal, then subtracts it. Used to remove a sloped or offset background from the raw signal.
Savitzky-Golay Filter	Polynomial fit over a sliding window	Noise filtering and smoothing method. Fits a low-degree polynomial to successive subsets of data points using least-squares regression (scipy.signal.savgol_filter). Reduces high-frequency noise while preserving peak shape and height. Configurable window size and polynomial order.
Fourier Transform Filter	$y' = \text{iFFT}[\text{top-}N\text{ components of FFT}(y)]$	Noise filtering and smoothing method. Transforms the signal to the frequency domain via FFT, retains only the $N$ most dominant frequency components (low-frequency = signal shape), and reconstructs the signal via inverse FFT. Removes high-frequency noise and oscillations. The number of retained frequencies $N$ is configurable.
Minimum Shift	$y' = y - \min(y)$	Translates the entire signal vertically so that its minimum value equals zero. Useful to remove any residual negative offset after baseline correction.
Negative Removal	$y' = \max(y, 0)$	Clips any remaining negative values to zero. Ensures the signal is non-negative before deconvolution, which assumes positive-definite Gaussian components.

---

Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page if you want to report a bug or request a feature.

---

Citation

If you use this software in your research, please cite it using the citation metadata provided in the CITATION.cff file.

---

License

This project is licensed under the Apache License 2.0 — see the LICENSE file for details.

Publisher: JELC