Metabolomic Tables: Peaks, Compounds, and Identification Confidence

What a Metabolomic Feature Table Records

Metabolomic assays often begin with instrument signals, not a list of known compounds. In a common liquid chromatography–mass spectrometry (LC–MS) workflow, molecules are separated over time and detected according to their mass-to-charge ratio. Software detects signal peaks and summarizes them as features. A feature is therefore an analytical signal defined by measured properties; it is not automatically a uniquely identified metabolite.

This distinction matters when interpreting a table. An intensity value describes the detected signal assigned to a feature under a particular assay and processing workflow. A compound name, if present, is an annotation based on evidence linking that signal to a possible chemical identity. The two columns answer different questions: “How much signal was recorded?” and “What might have produced it?”

In LC–MS, \(m/z\) is the mass-to-charge ratio of a detected ion, not necessarily the mass of a neutral molecule. Retention time is the time at which an ion elutes from the chromatographic separation, typically reported in seconds or minutes. Software may summarize the signal across the detected peak using peak height, integrated peak area, or another defined measure. Peak area is often used because it reflects signal across the peak’s width, but it remains an instrument-response measure.

For a chromatographic signal \(y(t)\) over a peak interval from \(t_0\) to \(t_1\), the idealized peak area is the integral of the signal above its estimated baseline. In practice, software estimates the baseline and approximates the area from discrete readings. For equally or unequally spaced readings, a common numerical approximation is the trapezoidal rule:

Here, \(y_i\) is the baseline-corrected signal at time \(t_i\), and \(A\) has units of signal multiplied by time. The calculation assumes the selected readings represent the peak and that the baseline and peak boundaries are reasonable. It does not convert signal into concentration by itself.

From a Signal Peak to a Compound Annotation

One compound can generate several detected signals. During ionization, a molecule may acquire or lose a proton, associate with a sodium or other ion, or form other adducts. Isotopes can also produce related peaks. Conversely, different compounds may produce ions with the same or very similar \(m/z\), and isomers may have matching masses. Chromatographic retention time and fragmentation patterns can help distinguish candidates, but no single property necessarily establishes a unique identity.

A detected feature is typically represented by a feature identifier, its measured properties, and sample-specific intensity values. An annotation may add a proposed compound name, a candidate database match, evidence such as tandem mass spectra, and an identification confidence level. These fields should remain conceptually separate: the feature ID identifies a processed signal, whereas the proposed compound name describes an interpretation of that signal.

The exact evidence available varies by platform and study. A database mass match alone is generally not equivalent to Level 1 identification: multiple compounds can share a molecular formula or similar mass. Likewise, a plausible biological pathway does not prove that a particular feature is a particular compound. Confidence should reflect analytical evidence, not how attractive a candidate seems biologically.

Worked Interpretations of Feature Records

Assumptions, Diagnostics, and Alternatives

Interpreting feature intensities assumes that the peak-detection and integration procedure has identified the intended signal rather than noise, a neighboring peak, or a baseline fluctuation. Comparing intensities across samples also requires care: extraction yield, ion suppression, instrument response, and run conditions can affect the signal. The size of an area is therefore most directly meaningful within the measurement process that produced it. A numerical value alone does not guarantee comparable biological amounts across different compounds or assays.

Useful diagnostics include inspecting extracted-ion chromatograms for representative features, checking whether peak boundaries include the signal but exclude neighboring peaks, reviewing blank samples for contamination, and checking quality-control samples for unstable or distorted signals. An annotation merits additional scrutiny if its proposed identity conflicts with the observed retention behavior, if its spectrum is weak or ambiguous, or if several plausible candidates remain. A feature that appears in blanks or whose peak shape is not credible should not be treated as a reliable biological signal merely because software assigned it a name.

Targeted quantitative LC–MS offers a different measurement goal from broad profiling. A targeted assay may use authentic standards, internal standards, calibration curves, and validated procedures to estimate concentrations for selected compounds. Broad profiling often emphasizes detecting many features, some of which remain putatively annotated or unknown. Nuclear magnetic resonance (NMR) provides another measurement approach, with different sensitivity and identification strengths. These approaches are not interchangeable: the suitable choice depends on the compounds of interest, desired quantitative accuracy, and available analytical evidence.

The central limitation is that feature detection and compound identification are separate tasks. A well-measured peak can remain chemically unknown; a plausible compound annotation can be attached to a signal whose intensity is not a calibrated concentration. Reporting both the measured feature properties and the evidence level makes those limits visible and supports sound downstream interpretation.

Common Pitfalls

• Treating every feature as one unique compound. Adducts and isotope peaks can produce multiple features from one compound, while distinct compounds can share similar analytical properties. Review evidence before equating feature counts with compound counts.
• Calling an intensity a concentration. Peak area is an instrument response. A concentration claim requires a suitable quantitative calibration and method.
• Reading confidence levels as probabilities. Level 2 does not mean a 2% or any other numerical probability of correctness. It describes the type of supporting evidence.
• Promoting a database match to a confirmed identity. Mass or library matches can narrow candidates but do not replace the required standard-based confirmation for Level 1.
• Using biological plausibility as identification evidence. A candidate’s role in a pathway may motivate follow-up, but it cannot by itself establish the chemical identity of a measured feature.