Comparative analysis of peak-detection techniques for comprehensive two-dimensional chromatography

doi:10.1016/j.chroma.2011.07.052

Journal of Chromatography A

Volume 1218, Issue 38, 23 September 2011, Pages 6792-6798

https://doi.org/10.1016/j.chroma.2011.07.052 Get rights and content

Abstract

Comprehensive two-dimensional gas chromatography (GC×GC) is a powerful technology for separating complex samples. The typical goal of GC×GC peak detection is to aggregate data points of analyte peaks based on their retention times and intensities. Two techniques commonly used for two-dimensional peak detection are the two-step algorithm and the watershed algorithm. A recent study [4] compared the performance of the two-step and watershed algorithms for GC×GC data with retention-time shifts in the second-column separations. In that analysis, the peak retention-time shifts were corrected while applying the two-step algorithm but the watershed algorithm was applied without shift correction. The results indicated that the watershed algorithm has a higher probability of erroneously splitting a single two-dimensional peak than the two-step approach. This paper reconsiders the analysis by comparing peak-detection performance for resolved peaks after correcting retention-time shifts for both the two-step and watershed algorithms. Simulations with wide-ranging conditions indicate that when shift correction is employed with both algorithms, the watershed algorithm detects resolved peaks with greater accuracy than the two-step method.

Introduction

Comprehensive two-dimensional gas chromatography (GC×GC) is a powerful technology for separating and analyzing compounds in complex samples. GC×GC data is processed to detect peaks and identify the associated compounds present in a sample. Typically, the goal of peak detection is to separately aggregate the data points belonging to each analyte peak. GC×GC peak detection popularly is performed by one of two approaches: the two-step algorithm [1] and the watershed algorithm [2], [3]. In the two-step algorithm, traditional one-dimensional (1D) peak detection is employed on each secondary chromatogram, then detected 1D peaks are merged to form two-dimensional (2D) peaks. The watershed algorithm performs peak detection by operating on 2D neighborhoods, i.e., in both retention-time dimensions simultaneously.

A recent study by Vivó-Truyols and Janssen [4] analyzed the effects of second-column retention-time shifts on the performance of 2D peak-detection techniques. Retention-time shift in consecutive secondary chromatograms in GC×GC may occur due to factors such as rapid temperature or pressure changes [5]. Rapid chromatographic changes can induce shifts that complicate data processing because all chromatographic peaks of a compound are expected to have the same retention-time. If the chromatography is rapidly varying (and cannot be improved to yield data without rapid peak shifts), then data processing and peak detection should account for retention-time shifts. Skov et al. [6] examined the nature and theory of retention-time shifts in GC×GC and described a method for shift correction based on cross-correlation for individual mass channels in adjacent secondary chromatograms. In the two-step algorithm, 1D peak merging typically accounts for retention-time shifts. For the watershed algorithm, retention-time shifts can be determined (e.g., with cross-correlation [6]) and then corrected either by aligning the data before peak detection or equivalently by adjusting the 2D neighborhoods of the watershed algorithm to account for shifts.

In their analysis, Vivó-Truyols and Janssen [4] compared the effects of retention-time shifts on both the two-step and watershed peak-detection algorithms. For the two-step approach, 1D peak merging accounted for retention-time shifts, whereas no shift corrections were made with the watershed algorithm. Their results indicated that watershed algorithm failed at a higher rate due to uncorrected second-dimension retention-time shifts. However, their comparison of peak-detection performance was confounded by accounting for retention-time shift in the two-step algorithm but not accounting for retention-time shift in the watershed algorithm.

This paper re-evaluates the performance of the two-step and watershed algorithms for retention-time shifts in the second-column separations when both employ shift correction. Simulated data is used to rigorously evaluate the peak-detection algorithms under controlled conditions with different levels of noise, retention-time shifts, and peak widths. The retention-time shifts in each simulated peak are corrected prior to both peak-detection algorithms for an unbiased comparison. These experiments demonstrate that when retention-time shift is corrected for both algorithms, the watershed algorithm detects peaks more accurately, over a wide range of conditions.

Section snippets

Two-step peak detection algorithm

The two-step algorithm builds on peak-detection methods used in traditional gas chromatography. In the first step, 1D peaks in each secondary chromatogram are detected by a 1D peak-detection algorithm. In the second step, a peak merging algorithm determines which of the detected 1D peaks should be merged to form 2D peaks [7].

The two-step algorithm typically uses two criteria to determine which 1D peaks should be merged: the overlap and unimodality criteria. The overlap criterion compares two

Overview of simulation

In the experiments described here, simulated 2D chromatographic peaks are used to rigorously compare both peak-detection algorithms under controlled conditions for different levels of noise, retention-time shifts, and peak widths. A single, resolved peak is simulated by interval sampling a 2D Gaussian function with parametric retention-time peak widths. Fig. 2(a) displays an example 2D peak as a series of 1D peaks. Each 1D second-column peak models a 1D chromatogram at a characteristic

Results

Table 1 compares results of both peak-detection algorithms for various parameter values, with 1000 test cases for each set of parameter values. In all cases shown in Table 1, the skew was −1, but, with skew correction, results for other levels of skew are not materially different. (Complete experimental test results for the simulations are reported as Supplementary Data.)

Table 1 lists the volume means and standard deviations for the signal and both peak-detection algorithms with different noise

Conclusion

A study by Vivó-Truyols and Janssen [4] discussed the probability of failure of the watershed algorithm for GC×GC data with varying second-dimension retention-time shifts. Their analysis compared the two-step and watershed peak-detection algorithms without accounting for the retention-time shifts in the watershed algorithm, whereas the shift was accounted for in the two-step algorithm. This caused the watershed algorithm to have a larger probability of failure than the two-step approach. This

Acknowledgements

This research was funded in part by the Nebraska Center for Energy Sciences Research at the University of Nebraska – Lincoln and the U.S. National Science Foundation under Award Number IIP-1013180.

References (21)

ReichenbachS. et al.
Chemom. Intell. Lab. Syst.
(2004)
Vivó-TruyolsG. et al.
J. Chromatogr. A
(2010)
RamosL. et al.
SkovT. et al.
J. Chromatogr. A
(2009)
PetersS. et al.
J. Chromatogr. A
(2007)
van der KliftE.J.C. et al.
J. Chromatogr. A
(2008)
ReichenbachS.
SinhaA.E. et al.
J. Chromatogr. A
(2004)
SinhaA.E. et al.
J. Chromatogr. A
(2004)
FragaC.G. et al.
J. Chromatogr. A
(2005)

There are more references available in the full text version of this article.

Cited by (29)

Critical comparison of background correction algorithms used in chromatography
2022, Analytica Chimica Acta
Citation Excerpt :
In such a case, or if the background is of a frequency very similar to that of the relevant signals, problems may occur with the interpretation of the data. For example, peak detection may be hindered, and errors in classification, discrimination, and, especially, quantification, may occur [1–7]. It is, therefore, desirable to perform baseline-drift correction and noise removal to ensure a correct interpretation of the data, unless peak detection can be performed in such a way that it is not hindered by the presence of noise and drift.
The objective of the present work was to make a quantitative and critical comparison of a number of drift and noise-removal algorithms, which were proven useful by other researchers, but which had never been compared on an equal basis. To make a rigorous and fair comparison, a data generation tool is developed in this work, which utilizes a library of experimental backgrounds, as well as peak shapes obtained from curve fitting on experimental data. Several different distribution functions are used, such as the log-normal, bi-Gaussian, exponentially convoluted Gaussian, exponentially modified Gaussian and modified Pearson VII distributions. The tool was used to create a set of hybrid (part experimental, part simulated) data, in which the background and all peak profiles and areas are known. This large data set (500 chromatograms) was analysed using seven different drift-correction and five different noise-removal algorithms (35 combinations). Root-mean square errors and absolute errors in peak area were determined and it was shown that in most cases the combination of sparsity-assisted signal smoothing and asymmetrically reweighted penalized least-squares resulted in the smallest errors for relatively low-noise signals. However, for noisier signals the combination of sparsity-assisted signal smoothing and a local minimum value approach to background correction resulted in lower absolute errors in peak area. The performance of correction algorithms was studied as a function of the density and coverage of peaks in the chromatogram, shape of the background signal, and noise levels. The developed data-generation tool is published along with this article, so as to allow similar studies with other simulated data sets and possibly other algorithms. The rigorous assessment of correction algorithms in this work may facilitate further automation of data-analysis workflows.
An automated system for predicting detection limit and precision profile from a chromatogram
2020, Journal of Chromatography A
Citation Excerpt :
This paper places more emphasis on noise than signal throughout the stages of research. As for signal, sophisticated strategies for peak detection can be found in the literature [25–31]. The uncertainty prediction of this paper still has room for improvement in peak detection, but it achieves the intended goal, within the domain of applicability, that a chromatogram is the only requirement for the DL and precision profile.
This paper presents a basic model of an automated system for predicting the detection limit and precision profile (plot of relative standard deviation (RSD) of measurements against concentration) in chromatography. The fundamental assumption is that the major source of response errors at low sample concentrations is background noise and at high concentrations, it is the volumes injected into an HPLC system by a sample injector. The noise is approximated by the mixed random processes of the first order autoregressive process AR(1) and white noise. The research procedures are: (1) the description of the standard deviation (SD) of measurements in terms of the parameters of the mixed random processes; (2) the algorithm for the parameter estimation of the mixed processes from actual background noise; (3) the mathematical distinction between noise and signal in a chromatogram. When compounds are chromatographically separated, each obtained signal is given the detection limit and precision profile on laboratory-made software. A file of a chromatogram is the only requirement for the theoretical prediction of measurement uncertainty and therefore the repeated measurements of real samples can be dispensed with. The theoretically predicted RSDs are verified by comparing them with the statistical RSDs obtained by repeated measurements. Signal shapes on noise are illustrated at the detection limit and quantitation limit, the signal-to-noise ratios of which are close to the widely adopted values, 3 and 10, respectively.
Introduction to Two-Dimensional Liquid Chromatography—Theory and Practice
2017, Handbook of Advanced Chromatography/Mass Spectrometry Techniques
Two-dimensional liquid chromatography (2D-LC) is becoming more widely adopted across a variety of industries to solve difficult separation problems. Increased availability and improvements in commercially available instruments in recent years are lowering the barrier to wider use. Users are increasingly focused on the development of powerful methods that can be deployed in both research and regulated laboratory environments. This chapter first establishes a framework for discussion by reviewing the major guiding principles for developing effective 2D-LC methods, and then goes on to discuss the most important advances in instrumentation, method development and optimization principles, and software over the past decade. Both new users and experts should find the chapter informative.
Peak detection methods for GC × GC: An overview
2016, TrAC - Trends in Analytical Chemistry
Citation Excerpt :
The probability of failure, such as e.g. oversegmentation, caused by differences in 2tr values between peaklets that belong to a single compound (e.g. due to temperature programming and instrumental variation) was studied in detail by Vivó-Truyols and Janssen [120] and was found to be 15–20% under normal conditions. In response, Latha et al. [121] showed that the rate of failure is much lower when shift correction is applied. However, this correction cannot be performed as an automated step in GCImage data processing.
The literature on methods for peak detection and integration for comprehensive two-dimensional gas chromatography (GC × GC) is reviewed with a focus on methods for the detection of individual peaks. These methods are of importance to make GC × GC generally applicable for identification and subsequent quantification of target compounds, and for non-target screening strategies, such as used in, e.g. food, biological, environmental and metabolomics studies, as well. There are three main groups of technique: methods based on 1D peak integration and subsequent automated clustering, multivariate methods and graphical methods. The principles, merits and demerits of each technique are discussed.
A simple multi-scale Gaussian smoothing-based strategy for automatic chromatographic peak extraction
2016, Journal of Chromatography A
Peak detection is a critical step in chromatographic data analysis. In the present work, we developed a multi-scale Gaussian smoothing-based strategy for accurate peak extraction. The strategy consisted of three stages: background drift correction, peak detection, and peak filtration. Background drift correction was implemented using a moving window strategy. The new peak detection method is a variant of the system used by the well-known MassSpecWavelet, i.e., chromatographic peaks are found at local maximum values under various smoothing window scales. Therefore, peaks can be detected through the ridge lines of maximum values under these window scales, and signals that are monotonously increased/decreased around the peak position could be treated as part of the peak. Instrumental noise was estimated after peak elimination, and a peak filtration strategy was performed to remove peaks with signal-to-noise ratios smaller than 3. The performance of our method was evaluated using two complex datasets. These datasets include essential oil samples for quality control obtained from gas chromatography and tobacco plant samples for metabolic profiling analysis obtained from gas chromatography coupled with mass spectrometry. Results confirmed the reasonability of the developed method.
Forensic potential of comprehensive two-dimensional gas chromatography
2016, TrAC - Trends in Analytical Chemistry
In this study, the application of comprehensive two-dimensional (2D) gas chromatography (GC × GC) in forensic science is reviewed. The peer-reviewed publications on the forensic use of GC × GC and 2D gas chromatography with mass spectrometric detection (GC × GC–MS) have been studied in detail, not only to provide an overview, but also to analyze and understand the potential and limitations of this relatively novel analytical technique in forensic investigations. In order to make comprehensive GC × GC a powerful tool in forensic laboratories involved in chemical analysis, further developments are needed with respect to data analysis. Easy-to-use and “fit-for-forensic purpose” data analysis and chemometric tools will allow forensic analytical chemists to explore and compare the comprehensive data to provide solutions for relevant questions in criminal investigations.

View all citing articles on Scopus

View full text

Comparative analysis of peak-detection techniques for comprehensive two-dimensional chromatography

Abstract

Introduction

Section snippets

Two-step peak detection algorithm

Overview of simulation

Results

Conclusion

Acknowledgements

Chemom. Intell. Lab. Syst.

J. Chromatogr. A

J. Chromatogr. A

J. Chromatogr. A

J. Chromatogr. A

J. Chromatogr. A

J. Chromatogr. A

J. Chromatogr. A