Limits...
Comparison of Several Methods of Chromatographic Baseline Removal with a New Approach Based on Quantile Regression.

Komsta L - Chromatographia (2011)

Bottom Line: It is compared with current methods based on polynomial fitting, spline fitting, LOESS, and Whittaker smoother, each with thresholding and reweighting approach.For curve flexibility selection in existing algorithms, a new method based on skewness of the residuals is successfully applied.The newly introduced methods could be preferred to visible better performance and short computational time.

View Article: PubMed Central - PubMed

Affiliation: Department of Medicinal Chemistry, Medical University of Lublin, Jaczewskiego 4, 20-090 Lublin, Poland.

ABSTRACT
The article is intended to introduce and discuss a new quantile regression method for baseline detrending of chromatographic signals. It is compared with current methods based on polynomial fitting, spline fitting, LOESS, and Whittaker smoother, each with thresholding and reweighting approach. For curve flexibility selection in existing algorithms, a new method based on skewness of the residuals is successfully applied. The computational efficiency of all approaches is also discussed. The newly introduced methods could be preferred to visible better performance and short computational time. The other algorithms behave in comparable way, and polynomial regression can be here preferred due to short computational time.

No MeSH data available.


Related in: MedlinePlus

Densitometric signal dataset: original (a), baseline filtered by quantile regression (b), baseline filtered by airPLS (c) and estimates of the baselines from quantile regression (d) and airPLS (e)
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3064906&req=5

Fig5: Densitometric signal dataset: original (a), baseline filtered by quantile regression (b), baseline filtered by airPLS (c) and estimates of the baselines from quantile regression (d) and airPLS (e)

Mentions: The study of quantile regression baseline estimation on experimental dataset was compared with Whittaker smoother approach implemented in airPLS package, after proper denoising of densitometric real signals [19]. The results are depicted in Figs. 5 and 6. The similarity between signals was depicted (Fig. 6) by Principal Component Analysis. The unprocessed signals (Fig. 6A) show very disturbed clustering, and similarity between several densitograms of the same oil is strongly affected by random variation of baseline. On the contrary, after baseline filtering (Fig. 6b and c) the densitograms of the same oil are very similar and whole dataset clusters visibly against the type of the essential oil. Therefore, it can be concluded that both algorithms perform well and the signal variability created by baseline drifts was almost removed. However, airPLS algorithm fits non-smooth baseline, and differences between samples of the same origin in PCA plot (Fig. 6b) are slightly larger. Moreover, several signals were not processed accurately with airPLS, and positive baseline is still visible. Quantile regression dealt with these issues. Comparing the computational time for this dataset (40 s for quantile regression and 350 s for airPLS, where the difference enhances with longer signals) the new quantile regression approach can be recommended.Fig. 5


Comparison of Several Methods of Chromatographic Baseline Removal with a New Approach Based on Quantile Regression.

Komsta L - Chromatographia (2011)

Densitometric signal dataset: original (a), baseline filtered by quantile regression (b), baseline filtered by airPLS (c) and estimates of the baselines from quantile regression (d) and airPLS (e)
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3064906&req=5

Fig5: Densitometric signal dataset: original (a), baseline filtered by quantile regression (b), baseline filtered by airPLS (c) and estimates of the baselines from quantile regression (d) and airPLS (e)
Mentions: The study of quantile regression baseline estimation on experimental dataset was compared with Whittaker smoother approach implemented in airPLS package, after proper denoising of densitometric real signals [19]. The results are depicted in Figs. 5 and 6. The similarity between signals was depicted (Fig. 6) by Principal Component Analysis. The unprocessed signals (Fig. 6A) show very disturbed clustering, and similarity between several densitograms of the same oil is strongly affected by random variation of baseline. On the contrary, after baseline filtering (Fig. 6b and c) the densitograms of the same oil are very similar and whole dataset clusters visibly against the type of the essential oil. Therefore, it can be concluded that both algorithms perform well and the signal variability created by baseline drifts was almost removed. However, airPLS algorithm fits non-smooth baseline, and differences between samples of the same origin in PCA plot (Fig. 6b) are slightly larger. Moreover, several signals were not processed accurately with airPLS, and positive baseline is still visible. Quantile regression dealt with these issues. Comparing the computational time for this dataset (40 s for quantile regression and 350 s for airPLS, where the difference enhances with longer signals) the new quantile regression approach can be recommended.Fig. 5

Bottom Line: It is compared with current methods based on polynomial fitting, spline fitting, LOESS, and Whittaker smoother, each with thresholding and reweighting approach.For curve flexibility selection in existing algorithms, a new method based on skewness of the residuals is successfully applied.The newly introduced methods could be preferred to visible better performance and short computational time.

View Article: PubMed Central - PubMed

Affiliation: Department of Medicinal Chemistry, Medical University of Lublin, Jaczewskiego 4, 20-090 Lublin, Poland.

ABSTRACT
The article is intended to introduce and discuss a new quantile regression method for baseline detrending of chromatographic signals. It is compared with current methods based on polynomial fitting, spline fitting, LOESS, and Whittaker smoother, each with thresholding and reweighting approach. For curve flexibility selection in existing algorithms, a new method based on skewness of the residuals is successfully applied. The computational efficiency of all approaches is also discussed. The newly introduced methods could be preferred to visible better performance and short computational time. The other algorithms behave in comparable way, and polynomial regression can be here preferred due to short computational time.

No MeSH data available.


Related in: MedlinePlus