Glossary
Limit of detection
The limit of detection defines the smallest amount that can be reliably told apart from nothing. It is a hard boundary of the measurement, and it is often confused with what a processing choice can and cannot recover.
Definition
The limit of detection is the lowest quantity of an analyte that can be reliably distinguished from the background at a stated confidence. It is commonly defined in terms of a signal to noise ratio or the standard deviation of blank measurements. In untargeted LC/MS the concept is practical rather than certified, because most features are unidentified and no calibration standard is available, but the idea still bounds discovery, since a compound present below the detection limit in every injection leaves no reproducible signal to recover. Distinguishing this hard instrument boundary from processing induced loss is important, because they have different remedies.
Worked example
In practice
A compound never sampled above the noise in any run cannot be recovered by processing, whereas a compound present above noise but discarded by a conservative threshold can be, which is a processing problem, not a detection limit.
What matters
Where this makes a difference
A boundary of the measurement
The detection limit is set by the instrument and method. No processing step lowers it, so it defines the outer edge of what any analysis can find.
Detection limit versus processing loss
Many lost features are above the detection limit but discarded by thresholds. That loss is recoverable at the feature layer, unlike a compound that was never measurable.
Why standards complicate it
Formal detection limits assume a calibrated analyte. In untargeted work most features are unknown, so the term is used as a practical bound rather than a certified value.
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.