Peak Shape Correlation Analysis

peak-shape-correlation-analysis

Summary

Filter isotopologue peaks by validating their chromatographic shape against the most abundant isotopologue in a cluster, removing peaks with Pearson correlation < 0.85 or intensity ratios deviating >30% from predicted values. This ensures only high-quality isotopologue clusters are retained for benchmark reliability assessment.

When to use

When extracting benchmark peaks from mzML files for multiple isotopologues of target molecules, after initial m/z and retention-time matching, to validate that detected isotopologue peaks exhibit consistent peak shape and expected abundance ratios before including them in a reliability assessment dataset.

When NOT to use

• When working with singly-charged or monoisotopic features only (no isotopologue clusters present).
• When isotopologue predictions are unavailable or unreliable for the target molecular formula.
• When the input intensity profiles lack sufficient chromatographic resolution (< 6 scans per peak) to compute meaningful correlation coefficients.

Inputs

• matched isotopologue peak clusters with m/z, retention time boundaries, and intensity profiles from extracted ion chromatograms
• enviPat-predicted isotopologue abundance ratios for target molecular formulas
• reference (most abundant) isotopologue intensity profile per cluster

Outputs

• validated isotopologue cluster records retained for benchmark dataset
• filtering flags indicating which isotopologues were removed (low correlation or degenerated ratio)
• benchmark peak table with quality-controlled isotopologue composition

How to apply

For each isotopologue cluster within a matched benchmark peak: (1) identify the theoretically most abundant isotopologue as the reference; (2) compute Pearson correlation coefficient between the intensity profile (across retention time scans) of each secondary isotopologue and the reference isotopologue; (3) remove any isotopologue with correlation < 0.85; (4) calculate the observed/predicted abundance ratio for each remaining isotopologue using enviPat predictions; (5) remove isotopologues where this ratio deviates by >30% from the predicted value. Retain only isotopologue clusters where at least the reference and one additional isotopologue pass both thresholds. This filtering ensures that abundance distortions caused by co-eluting features or detector artifacts do not corrupt the benchmark.

Related tools

• mzRAPP (Performs peak extraction, isotopologue matching, and applies peak-shape and ratio filters to generate benchmark datasets from mzML files) — https://github.com/YasinEl/mzRAPP
• enviPat (Predicts isotopologue m/z values and theoretical abundance ratios for target molecular formulas, enabling validation thresholds)
• R (Statistical computation environment for Pearson correlation coefficient calculation and filtering logic)

Examples

library(mzRAPP); callmzRAPP()  # Then in UI: load benchmark peaks → Configure isotopologue validation with peak_shape_correlation >= 0.85 and isotopologue_ratio_bias < 30% → Execute filtering

Evaluation signals

• Pearson correlation coefficient ≥ 0.85 between secondary and reference isotopologue intensity profiles across all retained isotopologues.
• Observed/predicted isotopologue ratio deviation ≤ 30% for all retained isotopologues.
• At least two isotopologues (reference + one secondary) pass both filters for each retained cluster.
• Benchmark dataset reports count and percentage of isotopologues removed due to low correlation vs. degenerated ratio.
• Comparison with NPP tool outputs shows improved degenerated IR metrics (3–20% range vs. baseline 28–53%).

Limitations

• Pearson correlation threshold (0.85) and ratio bias threshold (30%) are fixed; may not be optimal for all instrument types or sample matrices.
• Filtering assumes that peak shape is predominantly determined by chromatographic behavior; co-eluting peaks or in-source fragmentation can violate this assumption.
• Requires at least the reference isotopologue and one additional isotopologue to be detected; molecules lacking multiple detectable isotopologues are excluded entirely.
• Very low-abundance isotopologues (< 0.05 relative abundance) may fail filtering due to poor signal-to-noise ratio rather than true peak shape mismatch.

Evidence

• [methods] Peak shape correlation with most abundant isotopologue validation: "Sufficient quality of low peaks was ensured by removing isotopologues that do not satisfy criteria in peak shape (peak shape correlation with most abundant isotopologue)"
• [methods] Isotopologue ratio bias filtering threshold: "removing isotopologues that do not satisfy criteria in peak shape and abundance (Isotopologue ratio bias < 30%)"
• [readme] Correlation coefficient and dual-isotopologue requirement: "Isotopologue peaks with an area or height which is more than 30% off the predicted value or a Pearson Correlation coef < 0.85 (as compared to the highest isotopologue) are removed."
• [readme] Minimum isotopologue detection for benchmark inclusion: "Only isotopologues for which the theoretically most abundant and at least one additional isotopologue are found are considered for the final benchmark."
• [methods] Performance improvement from filtering: "In the Post Alignment box, we see that now about 93-99% of peaks have been detected, which is quite some improvement. Also, the proportion of degenerated IR decreased to 3-20%."