The effect of different baseline estimators on the limit of quantification in chromatography

Abstract

INTRODUCTIONChromatography is an important tool in analytical chemistry for identifying and quantifying the constituents of a chemical mixture. A typical chromatographic signal consists of a series of Gaussian-like peaks superimposed on a noisy background, (chromatogram = peaks + baseline + noise). The peaks represent the constituents of the chemical mixture while the background (baseline + noise) represents the physicochemical environment used to generate the chromatogram. The baseline is special in the sense that it is the notional zero from which all other points are measured and thus defines the chromatographic frame of reference.Although the baseline defines the chromatographic frame of reference, the baseline is obscured by both peaks and noise and needs to be estimated. A baseline can be estimated using interpolation, curve fitting or filtering.Interpolation methods estimate the value of a function between fixed points. When the magnitude of a peak is negligible (compared with the noise) at the detected start and end of the CCC SUMMARYThe ASC guidelines published in 1980 (ASC Committee on Environmental Improvement, 'Guidelines for data acquisition and data quality evaluation in environmental chemistry', Anal. Chem. 52, 2242-2249 (1980)) define the limit of quantification between a signal and its notional zero (baseline), i.e. the value of the signal without an analyte, as ten times the standard deviation of the signal's noise. If the value of the blank signal is estimated, the limit of quantification is dependent upon the variability of the estimated baseline rather than a simple multiple of the noise. It is shown that the limit of peak quantification (via numerical integration) of a Gaussian peak on an arbitrary background is strongly dependent upon the baseline estimator. In particular a peak should have a signal-to-noise ratio of approximately 60 : 1 using a linear baseline estimator and 10 : 1 for a cubic baseline estimator. Further, the expected standard deviation of peak parameters obtained via curve fitting is inflated by a factor of seven for a linear baseline estimator and two for a cubic baseline estimator.

Accuracy and Precision of Manual Baseline Determination

“…Noise is frequently defined as having zero mean 5,14 or implicitly treated as having zero mean. 8,[15][16][17] The results reported here show that inexperienced volunteers, unaware of the restrictions imposed by instrumentation, intuitively chose as baseline removal endpoint a condition with zero-mean noise.…”

An overview of signal processing issues in chemical sensing

“…In [28], the baseline is defined as "the portion of a detector record resulting from only eluant or carrier gas emerging from the column". Broader definitions exits, encompassing more deterministic components such as temperature fluctuations or even small peaks that cannot be easily distinguished from a notional or arbitrary zero level [29], serving as a reference for peak properties quantification (height, area). Removing a slow-varying, potentially monotone, trend ( Fig.…”

Chromatogram baseline estimation and denoising using sparsity (BEADS)

“…Several sources of uncertainties affect the quality and the performance of gas and liquid chromatography analysis [48,1]. As with many other analytical chemistry methods (including infrared or Raman spectra [6]), chromatogram measurements are often considered as a combination of peaks, background and noise [35]. The two latter terms are sometimes merged under different denominations: drift noise, baseline wander, or spectral continuum.…”