117828229

526 M. Daszykowski et al.

Step 4: Normalization

To take other deficiencies of the assessed herbal fingerprints into account, the last step of preprocessing was executed and the additional data trans-formations were applied. For chromatographic signals similar to those dis-cussed in this paper, different types of normalization can be considered. Probably the most frequently applied transformation is simple normalization of individual signals that enables correction of differences between in-jected sample volumes. This type of preprocessing was applied to both sets of sagę fingerprints. In addition to simple normalization, other types of normalization exist, e.g. scaling each signal to the largest peak observed in a signal or among all signals, the so-called Pareto scaling, vast scaling, etc. [37, 38].

To emphasize variability related to the variation among the sagę fingerprints, they were also centred before comparative analysis. In that way, constant terms are removed from the data and data variability is set about its mean.

A comprehensive overview of different data preprocessing approaches can be found elsewhere [37], and a detailed discussion is also given in Ref. [39], and in the references contained therein.

Exploratory Analysis of the Sagę (Salvia L.) Fingerprints

Investigated

After preprocessing of the fingerprints (removal of uninformative regions from the chromatograms, elimination of the background and noise compo-nents, alignment of the fingerprints, and normalization), it is possible to ex-plore differences among the samples. When data preprocessing is per-formed carefully, the undesired part of the data variation is removed. Com-parative analysis of the samples can now be focused solely on the informa-tive part of the data related to Chemical variability among the samples.

Many chemometric techniques are suitable for exploration of differences among samples. In our study, principal component analysis (PCA) was used to study differences among the chromatographic fingerprints of the sagę species.

The HS-GC-MS fingerprints represent the Chemical contents of the volatile fraction of Salina species. The first two principal components de-scribe ca 46.6% of the total data variance. Data compression is not very effi-cient, and the score plot presented in Fig. 5 reveals an inhomogeneous data structure. It is possible to distinguish three groups of the samples. The first group contains nine samples (1, 3, 4, 9, 10, 12, 14, 15, and 18), the second