Interference Correction • mrmhub

Tutorial

In targeted MRM assays, the signal of one transition can be perturbed by contributions from a co-eluting compound. The most common cause in lipidomics is overlap of natural-abundance isotopologues (M+1, M+2 from ¹³C, ²H, ¹⁵N) with the precursor or product window of an adjacent species. MRMhub implements a contribution-based subtraction (LICAR-style) (Gao et al. 2021) that removes the proportional signal of an interfering feature from a target feature on a per-injection basis.

This tutorial covers when to apply interference correction, how to define interference relationships in the feature annotation, and how to apply both batch and manual corrections.

Time ~15 min · Level Advanced · Prerequisites Basic workflow

1. When is correction needed?

Interference correction should be considered when:

blank injections show a non-zero signal in a feature that should be zero;
two features that should be biologically uncorrelated show a strong injection-to-injection correlation suggestive of cross-talk;
the theoretical M+1 / M+2 contribution from an adjacent species is large enough to bias the measurement (typically more than a few percent of the target signal at expected concentrations).

Small contributions (below ~1% of target signal) rarely justify correction; they should be documented but not necessarily subtracted, since the propagated uncertainty of the correction can exceed the bias removed.

2. Setup

library(mrmhub)

mexp <- readRDS("results/mexp_processed.rds")

Interference correction operates on raw feature intensities (feature_intensity). Apply it before ISTD normalisation, drift, and batch correction, so that those downstream steps use corrected raw signals.

3. Inspecting candidate pairs

plot_qc_interferences() provides a visual overview of features flagged as potentially interfering. For an exploratory check, retrieve the long-format dataset and correlate suspect features across injections:

d <- get_analyticaldata(mexp, annotated = TRUE)

d_wide <- d |>
  dplyr::select(analysis_id, feature_id, feature_intensity) |>
  tidyr::pivot_wider(names_from = feature_id, values_from = feature_intensity)

# Pearson correlation across QC injections for a suspected pair
suspect <- c("Cer 18:1;O2/16:0", "Cer 18:1;O2/16:0 M+2")
cor(d_wide[, suspect], use = "pairwise.complete.obs")

A correlation close to 1 between a species and a M+2 shoulder consistent with isotopologue overlap is a strong indicator that subtraction is warranted.

4. Batch correction via feature annotation

correct_interferences() reads the interference relationships defined in annot_features. Two columns are consulted:

Column	Description
`interference_feature_id`	`feature_id` of the feature contributing to the target signal
`interference_contribution`	Fraction (0–1) of the interfering feature’s intensity contributing to the target

Features without an interference assignment leave these columns empty (NA). When relationships are chained (A → B → C, i.e. C interferes with B which interferes with A), set sequential_correction = TRUE (default) so that downstream features are corrected first.

mexp <- correct_interferences(mexp,
                              variable = "feature_intensity",
                              sequential_correction = TRUE,
                              neg_to_na = FALSE)

For each affected feature the correction subtracts the interfering contribution:

$\text{Value}_\text{corrected} = \text{Value}_\text{target} - f_\text{contribution} \cdot \text{Value}_\text{interfering}$

Negative results are kept by default; setting neg_to_na = TRUE replaces them with NA and emits a warning. Circular dependencies in the interference graph (A → B → A) are detected and aborted with an informative error.

Required annotation column placement

Add interference_feature_id and interference_contribution to the annot_features table — the same table that defines feature classes and ISTD assignments. The MRMhub Metadata Organizer template includes these columns; the plain save_metadata_templates() output does as well in recent versions of the package.

5. Manual correction of a single pair

For one-off corrections, or when validating a contribution factor before adding it to the annotation, use correct_interference_manual(). Note that variable here is the actual column name in dataset (feature_intensity), not the short form accepted by the plotting functions.

mexp <- correct_interference_manual(
  mexp,
  variable                  = "feature_intensity",
  feature                   = "PC 32:0",
  interfering_feature       = "PC 32:0 | SM 36:1 M+3",
  interference_contribution = 0.0107,
  neg_to_na                 = FALSE,
  updated_feature_id        = NA
)

Setting updated_feature_id renames the corrected feature so the original and corrected values can coexist in the dataset under different IDs — useful when reporting both raw and corrected channels.

6. Verifying the correction

After correction the original raw values are preserved in feature_intensity_orig so that before/after comparisons remain available.

d <- get_analyticaldata(mexp, annotated = TRUE) |>
  dplyr::filter(feature_id == "PC 32:0") |>
  dplyr::select(analysis_id, qc_type,
                intensity_before = feature_intensity_orig,
                intensity_after  = feature_intensity) |>
  dplyr::mutate(pct_change = 100 * (intensity_after - intensity_before) / intensity_before)

summary(d$pct_change)

For blanks (SBLK/PBLK), residual signal after correction should approach zero. A non-zero median in blanks after correction often indicates that the contribution factor is underestimated.

7. Sourcing contribution factors

Contribution factors can be derived from three sources of varying confidence:

Theoretical natural-abundance isotopologue calculation — compute the M+n fraction from the neutral formula using enviPat (Loos et al. 2015) or comparable tools, then scale by any instrument-specific transmission differences across the two transitions.
Empirical measurement — inject the pure standard of the interfering species and measure the ratio of its signal in the target transition to its signal in its own transition. This captures instrument-specific effects (Q1/Q3 resolution, cross-talk, in-source fragmentation) that the theoretical calculation omits.
Published values — class-level interference tables published for lipid panels with shared precursor scans. Useful as a starting point but should be verified empirically when migrating methods between instruments.

Example: theoretical M+2 fraction with enviPat

library(enviPat)
data(isotopes)

# PC 32:0 — neutral formula C40H80NO8P
pattern <- isopattern(isotopes, "C40H80NO8P", charge = 1, threshold = 0.01)

# M+2 abundance relative to M+0
m_plus_2 <- pattern[[1]]
m_plus_2[, "abundance"] / max(m_plus_2[, "abundance"])

The reported M+2 fraction is the theoretical isotopologue abundance only. A real interference factor also depends on Q1/Q3 transmission, in-source fragmentation, and chromatographic resolution of the two species. Validate against blank or pure-standard injections before applying broadly.

8. Recommendations

Apply interference correction on raw feature_intensity before normalisation, drift, and batch correction.
Document every contribution factor and its provenance (theoretical, empirical, or published) in the annotation file or a companion notebook.
Re-validate factors after any method change (Q1/Q3 resolution, gradient, source temperature).
For small contributions (< 1% of target signal) the propagated uncertainty often exceeds the removed bias; document the relationship but consider leaving it uncorrected.
After correction, verify blank residual signal approaches zero.

Next steps

Drift and Batch Correction — apply after interference correction
Basic MRMhub Workflow — full processing pipeline
Design Decisions — rationale for the processing order
The MRMhubExperiment Data Object — how _orig postfixes preserve raw values

References

Gao, Liang, Shanshan Ji, Bo Burla, Markus R. Wenk, Federico Torta, and Amaury Cazenave-Gassiot. 2021. “LICAR: An Application for Isotopic Correction of Targeted Lipidomic Data Acquired with Class-Based Chromatographic Separations Using Multiple Reaction Monitoring.” Analytical Chemistry 93 (6): 3163–71. https://doi.org/10.1021/acs.analchem.0c04565.

Loos, Martin, Christian Gerber, Frederic Corona, Juliane Hollender, and Heinz Singer. 2015. “Nontarget Screening with High-Resolution Mass Spectrometry in the Environment: Ready to Go?” Environmental Science & Technology 49 (3): 1857–65. https://doi.org/10.1021/es5040179.