HolobiomicsLab Gaussian peakshape fitting and evaluation
Summary
Evaluates detected elution peaks against expected Gaussian shape to filter for chromatographic quality and reduce false positives. Applied after peak detection to retain only peaks with peakshape scores above a threshold (default > 0.5), ensuring detected features represent genuine metabolites rather than noise or artifacts.
When to use
After peak detection on mass track segments using find_peaks, when you need to distinguish genuine chromatographic peaks from noise-induced false positives or irregular shapes. Particularly important in high-intensity regions where prominence-based detection alone may capture non-Gaussian features or shoulders.
When NOT to use
- Input is already a feature table or pre-filtered peak list — use this skill during initial peak detection, not post-hoc validation.
- Data are from instruments with non-Gaussian elution profiles (e.g. gas chromatography with tailing or fronting, or highly irregular retention behavior) — Gaussian model assumptions may not hold.
- Peak detection has not yet been performed or apex positions are not reliably identified — peakshape evaluation requires a well-localized apex.
Inputs
- detected_peaks (list of peak objects with apex position, intensity window, m/z and retention time)
- mass_track_segment (region of interest: intensity array subset after baseline/noise subtraction)
- peakshape_threshold (numeric, default 0.5)
- snr_threshold (numeric, default 2)
Outputs
- filtered_peaks (peak objects passing peakshape and SNR criteria, with peakshape and SNR scores recorded)
- peak_quality_metrics (table of peakshape, SNR, and apex position for each retained peak)
How to apply
For each detected peak, fit a Gaussian model to the intensity values in a window around the apex position and calculate a peakshape metric (e.g., ratio of observed peak area to fitted Gaussian area, or R² of fit). Retain peaks only if peakshape exceeds the threshold (default 0.5). This filtering is applied in conjunction with SNR thresholds (default SNR > 2) and validates that the apex position lies within the segment boundaries. The rationale is that metabolite peaks follow Gaussian elution profiles in LC; deviations suggest either co-eluting compounds, baseline artifacts, or instrumental noise rather than a single resolved feature.
Related tools
- scipy.signal.find_peaks (Detects local maxima and prominence on mass track; output feeds into peakshape evaluation) — https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks.html
- asari.peaks.evaluate_gaussian_peak_on_intensity_list (Core function implementing Gaussian peakshape fitting and metric calculation) — https://github.com/shuzhao-li/asari
- asari.peaks.detect_evaluate_peaks_on_roi (Orchestrates peak detection and peakshape evaluation on each region of interest) — https://github.com/shuzhao-li/asari
- Python (numpy, scipy.optimize) (Underlying fitting and numerical optimization for Gaussian model)
Examples
from asari.peaks import evaluate_gaussian_peak_on_intensity_list; peakshape = evaluate_gaussian_peak_on_intensity_list(intensity_window, apex_idx); retained = peak if peakshape > 0.5 and snr > 2 else None
Evaluation signals
- Peakshape scores for retained peaks are >= 0.5 and SNR >= 2; rejected peaks are logged with their failure reason
- Apex positions of retained peaks fall strictly within segment boundaries (no boundary violations)
- Peakshape distribution (histogram or summary stats) is unimodal and concentrated in the 0.5–1.0 range, indicating consistent Gaussian fit quality
- Comparison of rejected vs. retained peaks shows that rejected peaks have visibly irregular or multi-modal intensity profiles around the apex
- Reproducibility check: same mass track and parameters re-processed yields identical peakshape scores and the same set of retained peaks
Limitations
- Assumes Gaussian elution profile; real peaks may be slightly tailed or fronted, especially at high intensity or under non-ideal chromatographic conditions.
- Peakshape threshold (default 0.5) is a heuristic; may require tuning for different instrument types, column chemistry, or metabolite classes.
- Co-eluting peaks or shoulder peaks may pass Gaussian fitting if the fit window is too narrow; requires integration with retention time alignment and mass separation.
- Very low-intensity peaks near noise floor may show artificially good peakshape due to stochastic noise that happens to fit a Gaussian; SNR threshold mitigates but does not eliminate this.
- Computational cost scales with number of detected peaks and window size; large feature tables may benefit from vectorized evaluation.
Evidence
- [other] Evaluate detected peaks for Gaussian peakshape, cSelectivity, and signal-to-noise ratio (SNR); retain peaks passing thresholds (default SNR >2, peakshape >0.5).: "Evaluate detected peaks for Gaussian peakshape, cSelectivity, and signal-to-noise ratio (SNR); retain peaks passing thresholds (default SNR >2, peakshape >0.5)."
- [other] Validation: verify that detected peaks have apex positions within segment boundaries, SNR and peakshape values within reported thresholds, and prominence values consistent with local signal characteristics.: "Validation: verify that detected peaks have apex positions within segment boundaries, SNR and peakshape values within reported thresholds, and prominence values consistent with local signal"
- [other] See peaks.evaluate_gaussian_peak_on_intensity_list: "See peaks.evaluate_gaussian_peak_on_intensity_list"