14 February / March 2018
Figure 2. New Workflow UI and Visualisation. The new workflow uses specialised detection and identification constraints to search for common and unique compounds. The result is displayed in a colour-labelled bubble plot that can be examined by analysts.
times and responses; and the samples and sample classes, for which relative measurements can be used to compare compositions. Such feature databases can be used for chemical fingerprinting, sample classification, chemical monitoring, sample clustering, and biomarker discovery. One important challenge is to develop data analysis and visualisation tools that can help select a few markers that can be used effectively for clustering and classifying multiple samples.
One common problem of marker selection is to detect unexpected compounds that appear in some samples but not others. A new workflow and associated tools are developed to allow analysts to detect common and unique compounds across many samples. This new workflow extends the Investigator framework with specialised detection and identification constraints that use chromatographic and mass spectral information to distinguish targeted compounds. In addition, new visualisation tools for multi-classification methods provide not only metric values, but also instructive predictions as to which features are effective for distinguishing samples. The workflow is demonstrated with two sample sets analysed by GCxGC coupled with quadrupole time-of- flight (Q-TOF) mass spectrometry.
Classical Statistics
Given a feature database extracted by the Investigator framework, classical statistical tools can be used for multiclass analysis to select constituents whose relative presence in samples are statistically related to the classes of the samples. For two sample classes, the Fisher Discriminant Ratio (FDR) is often used. FDR is the ratio of between-group variance to within-group variance [6, 7]. It can be used to assess pairwise class differences for the measure in each peak-region feature:
A large FDR or F value indicates a large separation of the class means relative to their within-class distributions. The direction of the change is indicated by the difference in means.
where FDR(x1 , x2 ) is the FDR for the sample
sets of measured values from two classes, x1
from Class 1 and x2
mean of sample values in xi variance of sample values in xi
from Class 2; μi ; and σi
2 is the . For multiple
sample classes, the F value is used to assess multiclass differences [7, 8]:
is the
where K is the number of classes, Ni number of sample values in xi
,j is the , N is the
number of sample values in all classes, μi is the mean of sample values in xi mean of all sample values, and xi value in xi
. For k=2 and N1 value are equal. =N2
, μ is the is the jth
, FDR and F
Although FDR and F value work well for traditional data classification analysis, they do not always accommodate practical requirements of chromatographic data analysis and chemical marker selection. For example, in practice, it may be expensive to collect multiple samples per class or to acquire multiple chromatographic runs for each sample. In situations with a single chromatographic run per sample class, FDR and F value cannot be computed (because they rely on within-class variance). Also, even if multiple chromatogram runs are available for each sample class, FDR or F value alone may not provide reliable predictions of commonality and uniqueness. For example, a compound feature with a high F value may be simply due to the response differences instead of identity differences across all samples. Thus, in order to detect unique compounds that appear in one sample class but not others or common compounds that appear in all samples, multiple attributes measured such as retention times, responses, and spectral information need to be used to cross-check the identities of chemical features.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68
