Solutions
Peak splitting in LC/MS feature detection
When a single compound is reported as two or more features, downstream statistics inherit the error. Peak splitting is one of the most common and most fixable sources of noise in a feature table.
The problem
Peak splitting in LC/MS feature detection
Peak splitting happens when the detection algorithm reports one analyte as several separate features. A compound that elutes as one chromatographic band gets carved into two or three narrow peaks, each with its own row in the feature table. The intensity of the real compound is divided across those rows, so every split feature looks weaker than the analyte actually is, and the same compound competes with itself in every statistical test that follows.
Why it happens
Where conventional workflows produce it
Conventional peak pickers fit an expected peak shape to the extracted ion chromatogram and cut on local minima. Real chromatography rarely gives a textbook shape. Fronting, tailing, saturation, and small dips at the apex all create false minima. When a run drifts or the mass tolerance is set tight, a single ion trace can break into fragments that the picker treats as independent. Aggressive smoothing hides the problem in one sample and exposes it in the next, so the split is inconsistent across a cohort, which is the worst case for alignment.
How Metablify addresses it
Evidence from the whole cohort
Metablify treats the cohort as evidence rather than scoring each sample in isolation. A feature that is real appears with consistent mass and retention behavior across many injections, so the platform amplifies that agreement and resolves fragments that belong to the same analyte back into one feature. Splits that are artifacts of one picker on one trace do not reproduce across the cohort and are suppressed. Because the decision is grounded in reproducibility, the merge is defensible rather than a cosmetic reshaping of peaks.
What changes
What you see in the output
The feature table carries one row per compound instead of several, intensities are whole rather than divided, and the same analyte stops appearing as multiple weakly correlated variables. Fold changes and multivariate models become easier to interpret because the variance that came from splitting is gone.
Honest limits
What this does not do
Metablify works on the data you already acquired, so it cannot recover a compound that coeluted completely and was never resolved at the instrument. Two isomers that share mass and retention will still require a chromatographic or ion mobility change to separate. The platform reduces detection artifacts; it does not add separation that the method did not provide.
What matters
Where this makes a difference
Split features divide real signal
When one compound becomes several rows, its intensity is spread across them, so every split feature understates the analyte and competes with itself in downstream statistics.
Cohort agreement resolves the merge
A real feature reproduces across injections. Metablify uses that agreement to merge fragments that belong together, so the correction is evidence based rather than a cosmetic reshaping of peaks.
Cleaner variance downstream
Removing splits reduces artificial variables and correlated noise, which makes fold changes and multivariate models easier to read and easier to trust.
Questions
Common questions
Is peak splitting the same as an isotope or adduct group?
No. Isotopes and adducts are real ions of the same compound at predictable mass offsets. Peak splitting is one ion trace broken into several features by the detection step. Both inflate a feature table, and both are handled at the feature layer, but the causes are different.
Can I fix splitting by smoothing more aggressively?
Heavier smoothing can merge a split in one sample while distorting the apex or hiding a low abundance neighbor in another. Because the effect varies across a cohort, tuning smoothing per sample trades one inconsistency for another. Resolving splits from cohort agreement is more stable.
Sources
- Smith et al., XCMS, Analytical Chemistry (2006) ↗Foundational peak detection method and its shape assumptions.
- Myers et al., peak detection evaluation, Analytical Chemistry (2017) ↗Comparison of peak detection behavior across tools.
Keep reading
Related
Prove it on your own data
Send a limited set of your existing LC/MS data and see how many additional real mass features are recovered against your current output.
Send a subset of your existing data and see how many split features collapse back to single compounds.