Full second-order chromatographic/spectrometric data matrices for automated sample identification and component analysis by non-data-reducing image analysis.
A data analysis method is proposed for identification and for confirmation of classification schemes, based on single- or multiple-wavelength chromatographic profiles. The proposed method works directly on the chromatographic data without data reduction procedures such as peak area or retention index calculation. Chromatographic matrices from analysis of previously identified samples are used for generating a reference chromatogram for each class, and unidentified samples are compared with all reference chromatograms by calculating a resemblance measure for each reference. Once the method is configured, subsequent sample identification is automatic. As an example of a further development, it is shown how the method allows identification of characteristic sample components by local similarity calculations thus finding common components within a given class as well as component differences between classes from the reference chromatograms. This feature is a valuable aid in selecting components for further analysis. The identification method is demonstrated on two data sets: 212 isolates from 41 food-borne Penicillium species and 61 isolates from 6 soil-borne Penicillium species. Both data sets yielded over 90% agreement with accepted classifications. The method is highly accurate and may be used on all sorts of chromatographic profiles. Characteristic component analysis yielded results in good agreement with existing knowledge of characteristic components, but also succeeded in identifying new components as being characteristic.