Peak evaluation metrics cselectivity snr gaussian fit

peak-evaluation-metrics-cSelectivity-SNR-gaussian-fit

Summary

Evaluate detected peaks against three orthogonal quality criteria—chromatographic selectivity (cSelectivity), signal-to-noise ratio (SNR > 2), and gaussian fit quality—to retain only peaks meeting shape and signal thresholds. This multi-metric filtering ensures peaks are reproducible, well-resolved from background, and conform to expected peak shape.

When to use

After scipy.signal.find_peaks has identified candidate peaks on a composite mass track segment, evaluate each peak to decide whether to retain it in the final feature table. Use this skill when you have detected peaks with coordinates (m/z, retention time) and need to filter by chromatographic selectivity, signal quality, and peak shape before reporting to downstream analysis or annotation.

When NOT to use

• Peak detection has not yet been performed (use scipy.signal.find_peaks first).
• Input data are already feature-level aggregates (feature table) rather than individual mass-track peaks; this skill operates on single-m/z peak objects, not across-sample features.
• The mass track has not been audited (audit_mass_track must be applied to establish baseline, noise, and detrending); SNR and selectivity calculations depend on those outputs.

Inputs

• list of detected peak coordinates (m/z, scan number/retention time index)
• intensity values from the segment containing the peak
• baseline and noise estimates for the mass track (from audit_mass_track)
• full mass track intensities (for selectivity and Gaussian fit evaluation)

Outputs

• filtered list of peaks meeting SNR > 2, cSelectivity, and gaussian fit thresholds
• peak metadata including SNR, cSelectivity, and gaussian fit goodness_fitting score
• binary pass/fail decision per peak

How to apply

For each detected peak, compute three independent metrics: (1) Gaussian fit quality using evaluate_gaussian_peak_on_intensity_list, which fits a Gaussian curve to the peak region and scores goodness of fit; (2) chromatographic selectivity (cSelectivity) using _peaks_cSelectivity_stats, quantifying how well the peak is resolved from neighboring signals in the mass track; (3) signal-to-noise ratio (SNR) by dividing the peak's intensity by the noise floor (computed from bottom-signal quartiles via compute_noise_by_flanks). Retain only peaks where SNR > 2, cSelectivity exceeds a threshold (article does not specify exact value, but selectivity is tracked as a key metric), and gaussian fit goodness is acceptable. The rationale is that these three metrics are independent and complementary: SNR ensures signal strength, cSelectivity ensures chromatographic isolation, and gaussian fit ensures the peak conforms to expected peak shape and is not an artifact.

Related tools

• scipy.signal.find_peaks (detects candidate peak locations before this evaluation step) — https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks.html
• peaks.evaluate_gaussian_peak_on_intensity_list (fits Gaussian curve to peak region and computes goodness_fitting metric) — https://github.com/shuzhao-li/asari
• peaks._peaks_cSelectivity_stats (computes chromatographic selectivity (cSelectivity) for each peak) — https://github.com/shuzhao-li/asari
• peaks.compute_noise_by_flanks (estimates noise floor from flanking regions to enable SNR calculation) — https://github.com/shuzhao-li/asari
• peaks.audit_mass_track (pre-processes mass track (baseline, detrending, smoothing) required for clean evaluation) — https://github.com/shuzhao-li/asari

Evaluation signals

• All retained peaks have SNR > 2 (SNR = peak intensity / noise floor estimate); check that no peak with SNR ≤ 2 is retained.
• cSelectivity metric is computed and reported for every peak; verify that a cSelectivity value appears in the output peak metadata and is within expected range [0, 1] or similar normalized scale.
• gaussian_fit quality score (goodness_fitting) is computed and peaks meeting threshold are explicitly marked; confirm that peaks with poor Gaussian fit are rejected.
• Peak evaluation functions are invoked exactly once per detected peak (not multiple times or skipped); trace that evaluate_gaussian_peak_on_intensity_list, _peaks_cSelectivity_stats, and SNR calculation are called for each candidate.
• Final feature table includes cSelectivity, SNR, and gaussian fit columns; spot-check a sample of peaks to verify reasonable values (e.g., SNR > 2, cSelectivity not at extremes for real peaks).

Limitations

• Exact cSelectivity threshold value is not specified in the article; practitioners must calibrate this threshold empirically or use asari defaults.
• SNR calculation depends on accurate noise estimation via compute_noise_by_flanks, which may fail or underestimate noise in crowded or noisy mass tracks.
• Gaussian fit evaluation assumes peaks follow a Gaussian shape; highly asymmetric or multiply-charged peaks may fail this criterion even if they are valid metabolite peaks.
• The three metrics (SNR, cSelectivity, gaussian fit) are independent; a peak may fail one criterion while passing others—the article does not specify how to handle partial failures or whether weighted aggregation is used.
• Peak evaluation is applied per mass track segment; isolated peaks or peaks in sparse segments may have unreliable selectivity or fit estimates due to limited neighboring signal context.

Evidence

• [methods] Evaluate detected peaks for gaussian fit (goodness_fitting), chromatographic selectivity (cSelectivity), and signal-to-noise ratio (SNR > 2), retaining only peaks meeting thresholds.: "Evaluate detected peaks for gaussian fit (goodness_fitting), chromatographic selectivity (cSelectivity), and signal-to-noise ratio (SNR > 2), retaining only peaks meeting thresholds."
• [methods] See peaks.evaluate_gaussian_peak_on_intensity_list: "See peaks.evaluate_gaussian_peak_on_intensity_list"
• [methods] See peaks._peaks_cSelectivity_stats: "See peaks._peaks_cSelectivity_stats"
• [methods] See peaks.compute_noise_by_flanks.: "See peaks.compute_noise_by_flanks."
• [intro] Tracking peak quality, selectiviy metrics on m/z, chromatography and annotation databases: "Tracking peak quality, selectiviy metrics on m/z, chromatography and annotation databases"