Summary of Results: Chemometrics with PDMS Data Spectral and Structural Similarity |
|
A prominent task in the
pre-evaluation of data on board of COSIMA will be the comparison of a measured
spectrum with previously measured spectra. The result will influence the
decision whether a sample site with a "new composition" has been
found (and the measurements should be continued) or not.
Therefore the applied spectra similarity
criterion should
well reflect chemical structure similarities.
Preliminary results about a new method for
calculating mass spectral similarities are briefly presented here.
Mass spectral similarity, s, was simply defined by the correlation
coefficient as being also used in many library search systems.
s = S (yiA yiB) / [ S (yiA) S (yiB)
]0.5
yiA and yiB variables i in the two compared spectra A and B.
Three types of variables, yi , have been compared:
(a) peak
intensities,
(b) spectral features,
(c) a set of
multivariate classification
models.
The similarities of chemical structures
in the hitlist and the chemical structures of used test compounds have been
measured by the Tanimoto index (using 135 binary molecular descriptors).
PRELIMINARY RESULTS
J In general the structural similarity
of the hitlist compounds is improved if spectral features are used to
calculate the spectral similarity instead of peak intensities.
J For the tested PDMS data a
further improvement is obtained by using the more complex responses of
multivariate models as variables for the calculation of spectral
similarities.
Spectral features and responses from
multivariate models can be considered as powerful "numerical sensors"
for the chemical structure. Electron impact (EI) mass spectra showed the same trend.
Other
subjects
|
[ Aims | People | Results | Presentations | Pictures | ROSETTA | COSIMA Instrument
| Comet
Wirtanen | Literature
] |
Last update 2000-12-03 |