Calibration Residual Calculation

calibration-residual-calculation

Summary

Computation of mass calibration coefficients and residuals from matched reference m/z points using least-squares fitting or polynomial regression. This skill assesses the quality and accuracy of mass spectrum calibration by quantifying the deviation between theoretical and observed m/z values.

When to use

Apply this skill after a mass spectrum has been matched against a reference m/z file (e.g., SRFA.ref) and a sufficient number of calibration points (≥5) have been identified within a given PPM tolerance window. Use it to finalize the calibration procedure and produce quantitative measures of calibration fit quality and point-specific errors.

When NOT to use

• Fewer than 5 matched reference points are available — the procedure has not identified sufficient calibration anchors; return to PPM window widening or data quality assessment instead.
• Reference m/z file is absent or incompatible with the mass spectrum format.
• Spectrum is already fully calibrated and residuals are not required for downstream analysis.

Inputs

• mass spectrum data object (e.g., from CoreMS Bruker .d or ThermoFisher .raw file)
• reference m/z file (e.g., SRFA.ref with theoretical m/z values)
• list of matched reference–observed m/z pairs with count ≥5
• PPM tolerance threshold used for matching

Outputs

• calibration coefficients (linear, quadratic, or Ledford equation parameters)
• residual values for each matched calibration point
• calibration quality metrics (point count, fit statistics)
• structured output record (serialized to file: .csv, .txt, .hdf5, or pandas DataFrame)

How to apply

After loading the mass spectrum (e.g., from ESI_NEG_SRFA.d) and matching reference m/z values against spectrum peaks, collect the matched reference–observed m/z pairs. Apply least-squares fitting or equivalent polynomial regression (linear, quadratic, or Ledford equation) to compute calibration coefficients that transform observed m/z to calibrated m/z. Calculate residuals as the difference between observed and fitted m/z values for each matched point. Serialize the results (point count, coefficients, residuals) into a structured record and write to output file. The quality of the calibration is reflected in the magnitude and distribution of residuals; smaller residuals and higher point counts indicate better calibration fidelity.

Related tools

• CoreMS (Mass spectrometry data encapsulation and calibration framework; provides spectrum loading, reference m/z matching, and polynomial regression/coefficient calculation) — https://github.com/EMSL-Computing/CoreMS
• numpy (Numerical computing for polynomial fitting and least-squares regression)
• pandas (Data frame construction and serialization of calibration results)
• Docker (Containerization for reproducible calibration workflow execution)

Examples

from corems.encapsulation.factory.parameters import MSParameters; from corems.mass_spectrum import MassSpectrum; ms = MassSpectrum(file_path='tests/tests_data/ftms/ESI_NEG_SRFA.d'); ref_mz = load_reference_mz('tests/tests_data/ftms/SRFA.ref'); matched_pairs = match_reference_mz(ms, ref_mz, ppm_tolerance=5.0); coeffs, residuals = compute_calibration(matched_pairs, method='linear'); output_df = pd.DataFrame({'coefficients': coeffs, 'residuals': residuals, 'point_count': len(matched_pairs)}); output_df.to_csv('calibration_results.csv')

Evaluation signals

• Point count matches the number of matched reference m/z values and is ≥5.
• Calibration coefficients are real numbers and have expected sign and magnitude relative to the PPM tolerance window used.
• Residuals are calculated for all matched points; the mean absolute residual should be smaller than the PPM window width used for matching.
• Output file is properly formatted (valid .csv, .txt, .hdf5, or .pkl) and contains all required fields: point count, coefficients, residuals.
• Residual distribution is approximately centered near zero with no systematic bias (e.g., large positive or negative skew) that would indicate a poor model fit.

Limitations

• Calibration accuracy depends entirely on the quality and completeness of the reference m/z file; errors or outliers in the reference data will propagate to coefficients and residuals.
• Polynomial order (linear vs. quadratic vs. Ledford) must be chosen appropriately for the mass spectrometer's m/z error behavior; mismatched order will yield poor residual distributions.
• With fewer than 5 calibration points, polynomial fitting becomes unstable and residuals are unreliable; the procedure requires prior successful PPM window widening to ensure sufficient matches.
• Residuals reflect calibration fit quality for the specific mass range covered by matched reference points; m/z values outside this range may have larger uncalibrated errors.

Evidence

• [other] When fewer than 5 calibration points are found within a given PPM error threshold, the procedure iteratively widens the threshold from ±1.0 ppm to ±1.5, ±3, ±5, ±7, and ±10 ppm until sufficient reference m/z matches are located.: "When fewer than 5 calibration points are found within a given PPM error threshold, the procedure iteratively widens the threshold from ±1.0 ppm to ±1.5, ±3, ±5, ±7, and ±10 ppm until sufficient"
• [other] Compute calibration coefficients and residuals from the matched points using least-squares fitting or equivalent polynomial regression.: "Compute calibration coefficients and residuals from the matched points using least-squares fitting or equivalent polynomial regression."
• [readme] Frequency and m/z domain calibration functions: LedFord equation, Linear equation, Quadratic equation: "Frequency and m/z domain calibration functions: LedFord equation, Linear equation, Quadratic equation"
• [other] Serialize the results (point count, coefficients, residuals) into a structured record and write to output file.: "Serialize the results (point count, coefficients, residuals) into a structured record and write to output file."
• [readme] Self-containing Hierarchical Data Format (.hdf5) including raw data and time-series data-point for processed data-sets with all associated metadata stored as json attributes: "Self-containing Hierarchical Data Format (.hdf5) including raw data and time-series data-point for processed data-sets with all associated metadata stored as json attributes"