Noise Smoothing Chromatographic Signals

noise-smoothing-chromatographic-signals

Summary

Apply smoothing filters to LC-HRMS chromatographic signals (retention time × m/z matrices) to reduce noise prior to peak detection. This preprocessing step is essential for accurate gradient-descent peak localization and feature extraction in untargeted metabolomics workflows.

When to use

When you have raw LC-HRMS profile-mode data (rt × mz intensity matrices) and need to detect chromatographic peaks using gradient-descent or local-maxima algorithms. The smoothing step is mandatory before peak border and center estimation to reduce false positives from background noise and instrument artifacts.

When NOT to use

• If you already have centroided or deconvoluted peak lists (not profile-mode raw data)
• If your LC-HRMS signal-to-noise ratio is already very high (e.g., SIM or MRM mode with minimal baseline noise); smoothing may become unnecessary and risk peak distortion
• If you are working with already-feature-extracted data or vendor-processed peak tables

Inputs

• raw LC-HRMS chromatogram data (retention time × m/z intensity matrix, typically NetCDF or mzML format)
• training chromatogram dataset

Outputs

• smoothed chromatographic signal (rt × mz matrix with attenuated noise)
• intermediate representation ready for gradient-descent peak detection

How to apply

Load your training chromatogram data in retention time × m/z matrix format. Apply a smoothing filter (specific kernel and window size are implementation-dependent; PeakBot's approach is designed for two-dimensional LC-HRMS data) to attenuate high-frequency noise while preserving peak shape integrity. The smoothed signal is then fed directly into a gradient-descent algorithm to locate chromatographic peak maxima and estimate peak borders and centers. The choice of smoothing strength should balance noise reduction against peak shape distortion; too aggressive smoothing will broaden or merge nearby peaks, while insufficient smoothing will leave noise-driven false maxima. Verify post-smoothing by visual inspection of a subset of chromatograms to confirm peaks remain well-resolved.

Related tools

• PeakBot (Python package that implements smoothing as a preprocessing step before gradient-descent peak detection in LC-HRMS chromatograms) — https://github.com/christophuv/PeakBot
• TensorFlow (Machine-learning framework used by PeakBot for CNN-based peak classification after smoothing and peak detection) — https://www.tensorflow.org/

Examples

python quickExample_GPU.py  # PeakBot example that includes smoothing as part of the training pipeline; or manually: from peakbot import preprocessing; smoothed_chromatogram = preprocessing.smooth_chromatogram(raw_rt_mz_matrix, kernel_size=5)

Evaluation signals

• Visual inspection: smoothed chromatogram shows reduced high-frequency noise while preserving peak maxima and shoulders
• Peak detection consistency: gradient-descent algorithm successfully locates expected number of local maxima without spurious noise-driven peaks
• Peak shape integrity: FWHM and peak asymmetry metrics of known reference peaks remain within ±10% of theoretical values post-smoothing
• Matching success rate: when matched against user-defined reference list, smoothed peaks exhibit rt and m/z alignment error < ±1 scan / ±5 ppm
• Signal recovery: intensity of smoothed peaks should not deviate >5% from raw peak area when integrated across the smoothed window

Limitations

• Smoothing strength is not quantitatively parameterized in the README; users must empirically validate for their instrument and sample type
• Two-dimensional smoothing (across both rt and m/z dimensions) may blur isomeric peak doublets or co-eluting features if applied too aggressively
• Performance depends on LC-HRMS data quality; severely noisy or poorly-calibrated instruments may require manual parameter tuning or preprocessing steps before smoothing
• Smoothing alone does not resolve true co-elution; downstream peak matching and CNN classification are required to distinguish overlapping peaks

Evidence

• [readme] searching for chromatographic peaks using a smoothing and gradient-descend algorithm: "searching for chromatographic peaks using a smoothing and gradient-descend algorithm"
• [readme] uses local-maxima in the LC-HRMS dataset each of which is then exported as a standarized two-dimensional area (rt x mz): "uses local-maxima in the LC-HRMS dataset each of which is then exported as a standarized two-dimensional area (rt x mz)"
• [readme] The peaks' borders and centers are also estimated in this step: "The peaks' borders and centers are also estimated in this step"
• [other] Load training chromatogram data (retention time × m/z matrix format). Apply smoothing filter to the chromatographic signal to reduce noise.: "Load training chromatogram data (retention time × m/z matrix format). Apply smoothing filter to the chromatographic signal to reduce noise."
• [readme] matched with a user-defined reference list (the ground-truth; isolated single chromatographic peaks) with the aim of using the same chromatographic peak but from different samples: "matched with a user-defined reference list (the ground-truth; isolated single chromatographic peaks)"