Glossary
Feature detection
Feature detection is the step that decides what counts as signal. Every later result is built on its output, so its false positives and false negatives set the boundaries of what a study can find.
Definition
Feature detection is the process of finding real ion signals in raw LC/MS data and distinguishing them from background noise. It typically extracts ion chromatograms within a mass tolerance, locates chromatographic peaks, and records each as a feature with an m/z, a retention time, and an intensity. The step must balance sensitivity against specificity, because a threshold that admits weak real signals also admits noise, and a threshold that excludes noise also discards real low abundance features. The choices made here define both the size and the reliability of the resulting feature table.
Worked example
In practice
Lowering an intensity threshold to catch a faint metabolite can double the feature count while filling the table with noise, which shows why detection is a balance rather than a single setting.
What matters
Where this makes a difference
The sensitivity and specificity balance
Detection lives on a tradeoff. Admitting weak signal raises recovery and noise together, so the discriminant used to separate them, not the threshold alone, determines quality.
Common failure modes
Peak splitting, missed low abundance features, and noise admitted as features are the recurring failures. Each has a different cause and a different fix at the feature layer.
Reproducibility as evidence
A feature that recurs across many injections is unlikely to be noise. Using cohort agreement as the discriminant is how Metablify raises recovery without simply lowering a threshold.
Keep reading
Related
Want to see the method on real data?
Read what a dataset assessment involves and what you get back.
Grounded in the same first principles that Metablify is built on.