Spectral Peak Alignment Across Pixels

spectral-peak-alignment-across-pixels

Summary

Aligns detected mass spectrometry peaks across all pixels in an imaging dataset to establish consistent m/z values for downstream feature-level analysis. This is a prerequisite for identifying peaks present in a sufficient fraction of pixels and for statistical analysis of imaging features.

When to use

After peak picking across individual spectra in an MSImagingExperiment, when you need to harmonize peak m/z positions across pixels to account for small shifts in peak location due to instrumental drift, calibration differences, or natural variation. Use this before subsetting to high-frequency peaks or before statistical analysis that assumes consistent peak identity across the imaging array.

When NOT to use

• Input peaks are already aligned or normalized to a reference peak list—alignment would be redundant.
• Working with a single spectrum or single pixel—alignment requires variation across multiple spectra to be meaningful.
• Peak positions are intentionally kept per-pixel for drift analysis or temporal tracking rather than harmonized for statistical analysis.

Inputs

• MSImagingExperiment with detected peaks in featureData (from peakPick())
• Spectra matrix with per-pixel peak detections

Outputs

• MSImagingExperiment with aligned peaks in featureData
• Unified peak m/z feature space across all pixels
• Peak frequency information (presence/absence per pixel)

How to apply

Load the MSImagingExperiment containing detected peaks (typically output from peakPick()). Apply peakAlign() to align peaks detected across all spectra in the imaging dataset—this function registers peak m/z values from different pixels to a common set of peak positions, reducing m/z jitter across the array. The alignment establishes a unified feature space where each aligned peak represents the same m/z across all pixels. After alignment, use subsetFeatures() with a frequency threshold (e.g., > 0.1 to retain peaks in >10% of pixels) to filter out spurious or rare peaks. Verify success by confirming the resulting featureData contains the expected number of aligned peaks and that their m/z values are stable across pixels.

Related tools

• Cardinal 3.6 (Provides peakAlign() function for aligning peak m/z positions across pixels and subsetFeatures() for frequency-based peak filtering) — https://github.com/kuwisdelu/Cardinal
• R (Runtime environment for executing Cardinal functions and MSImagingExperiment operations)
• BiocParallel (Optional parallel processing support for peakAlign() across large imaging datasets)

Examples

library(Cardinal); mse <- peakAlign(mse); mse_filtered <- subsetFeatures(mse, mz ~ frequency > 0.1)

Evaluation signals

• The resulting MSImagingExperiment featureData contains a consistent set of aligned peaks (e.g., 37 peaks) present across multiple pixels.
• Peak m/z values in featureData show reduced variance compared to pre-alignment per-pixel detections, indicating successful harmonization.
• Frequency counts in feature metadata indicate each peak is present in a reasonable fraction of pixels (e.g., ≥10% for common peaks).
• Subsetting to peaks with frequency > threshold retains only peaks meeting the frequency criterion; no spurious single-pixel peaks remain in the filtered result.
• Manual inspection or plotting of aligned peaks across pixel coordinates confirms visual alignment without m/z drift artifacts.

Limitations

• Peak alignment quality depends on sufficient peak overlap across pixels; sparse or highly localized spatial features may not align robustly.
• The choice of frequency threshold (e.g., > 0.1) is user-dependent and affects the final feature count; no universal optimal threshold is provided in the article.
• Alignment assumes peaks are real chemical species; false positives from noise in individual pixels may introduce spurious aligned features if peak-picking thresholds are too lenient.
• Large m/z shifts or systematic calibration drift across the imaging array may exceed alignment tolerance, resulting in multiple m/z entries for a single true peak.

Evidence

• [other] After filtering to peaks with frequencies > 0.1, the resulting MSImagingExperiment contains exactly 37 peaks in featureData.: "After filtering to peaks with frequencies > 0.1, the resulting MSImagingExperiment contains exactly 37 peaks in featureData."
• [other] Apply peakAlign() to align detected peaks across all spectra in the imaging dataset.: "Apply peakAlign() to align detected peaks across all spectra in the imaging dataset."
• [other] Subset features using subsetFeatures() with a frequency threshold of > 0.1 to retain only peaks present in >10% of pixels.: "Subset features using subsetFeatures() with a frequency threshold of > 0.1 to retain only peaks present in >10% of pixels."
• [intro] New spectral alignment methods in recalibrate(): Local maxima-based alignment using local regression, Dynamic time warping, Correlation optimized warping: "New spectral alignment methods in recalibrate(): Local maxima-based alignment using local regression, Dynamic time warping, Correlation optimized warping"
• [intro] Parallel processing support via the BiocParallel package for all pre-processing methods and any statistical analysis methods with a BPPARAM option: "Parallel processing support via the BiocParallel package for all pre-processing methods and any statistical analysis methods with a BPPARAM option"