331
12 Raman Spectra of Solid Aminoacids: Spectral Correlation Analysis …
the spectrometer sensitivity was calibrated with a Bentham CL2 spectral irradiance
standard lamp.
to prevent orientational dependence of Raman signals on the polarization of the
laser beam, the crystalline samples were carefully ground and dispersed on microscopic glasses. At least 100 point maps of Raman spectra were recorded for each
sample and then averaged. this procedure was repeated for all four laser lines. to
prevent the influence of long term laser fluctuations on the relative intensities of all
amino acids spectra, the laser power on the sample was monitored by measuring the
intensity of the 520.6 cm
−1
band of crystalline silicon.
Basics of statistical analysis the simplest method of spectral analysis is peak
identification directly from a graph. We identify peaks by taking their positions and
intensities and checking in a related database if they fit to some peaks of the reference spectrum of a substance. If the number of substances in the database increases, we take into account more peaks to make the method more specific. the next
step is to consider spectra as continuous functions and look in the database for a
spectrum most similar to the analysed one. As a standard measure of comparison for
this task, one can choose the Pearson correlation coefficient, the Euclidean distance
or the angle (dot product) between the spectra. It is obvious that this kind of analysis is well suited for computers, because doing it manually can be very tedious,
especially when the number of substances in a database, together with the number
of peaks to be considered, is growing. As a consequence, it is of no surprise that
computers have been used in this domain for a long time. For instance, in [5] the
authors identified single IR spectra by the Pearson correlation coefficient. they
varied the wavenumber range and the number of data points to find the optimum
values for this task. they also checked the stability of the method by shifting the
wavenumber scale and by perturbing the intensities. these steps are necessary to
account for the finite accuracy and reproducibility of the spectrometer, as well as for
a possible presence of impurities in the sample.
In case of the spectrum of a mixture the identification is more complicated, even
for non-interacting components, because in this simplest, ideal case the intensities
are sums of component intensities. one may hope that for some spectral regions the
intensities are due mostly to one component, i.e., this component is dominant in the
mixture spectrum for a particular spectral interval. In an identification algorithm,
one might look for a component spectrum most similar to the mixture spectrum,
then subtract the component spectrum and search for successive components by
comparing their spectra with the difference spectrum. this procedure stops when
some kind of stop condition is satisfied. Several relevant parameters must be defined, such as the tolerance with which we take the peak positions and intensities. If the algorithm is based on the identification of peaks, one must decide how
many peaks should be taken into account for the mixture and component spectra.
In the case of treating spectra as continuous functions, one must define the width
of a wavenumber interval to consider. the problem of appropriate interval choice
for matching spectra with a correlation function was addressed in [6]. the authors
presented a detailed analysis of simulated spectra. they varied peak positions and
12 Raman Spectra of Solid Aminoacids: Spectral Correlation Analysis …
the spectrometer sensitivity was calibrated with a Bentham CL2 spectral irradiance
standard lamp.
to prevent orientational dependence of Raman signals on the polarization of the
laser beam, the crystalline samples were carefully ground and dispersed on microscopic glasses. At least 100 point maps of Raman spectra were recorded for each
sample and then averaged. this procedure was repeated for all four laser lines. to
prevent the influence of long term laser fluctuations on the relative intensities of all
amino acids spectra, the laser power on the sample was monitored by measuring the
intensity of the 520.6 cm
−1
band of crystalline silicon.
Basics of statistical analysis the simplest method of spectral analysis is peak
identification directly from a graph. We identify peaks by taking their positions and
intensities and checking in a related database if they fit to some peaks of the reference spectrum of a substance. If the number of substances in the database increases, we take into account more peaks to make the method more specific. the next
step is to consider spectra as continuous functions and look in the database for a
spectrum most similar to the analysed one. As a standard measure of comparison for
this task, one can choose the Pearson correlation coefficient, the Euclidean distance
or the angle (dot product) between the spectra. It is obvious that this kind of analysis is well suited for computers, because doing it manually can be very tedious,
especially when the number of substances in a database, together with the number
of peaks to be considered, is growing. As a consequence, it is of no surprise that
computers have been used in this domain for a long time. For instance, in [5] the
authors identified single IR spectra by the Pearson correlation coefficient. they
varied the wavenumber range and the number of data points to find the optimum
values for this task. they also checked the stability of the method by shifting the
wavenumber scale and by perturbing the intensities. these steps are necessary to
account for the finite accuracy and reproducibility of the spectrometer, as well as for
a possible presence of impurities in the sample.
In case of the spectrum of a mixture the identification is more complicated, even
for non-interacting components, because in this simplest, ideal case the intensities
are sums of component intensities. one may hope that for some spectral regions the
intensities are due mostly to one component, i.e., this component is dominant in the
mixture spectrum for a particular spectral interval. In an identification algorithm,
one might look for a component spectrum most similar to the mixture spectrum,
then subtract the component spectrum and search for successive components by
comparing their spectra with the difference spectrum. this procedure stops when
some kind of stop condition is satisfied. Several relevant parameters must be defined, such as the tolerance with which we take the peak positions and intensities. If the algorithm is based on the identification of peaks, one must decide how
many peaks should be taken into account for the mixture and component spectra.
In the case of treating spectra as continuous functions, one must define the width
of a wavenumber interval to consider. the problem of appropriate interval choice
for matching spectra with a correlation function was addressed in [6]. the authors
presented a detailed analysis of simulated spectra. they varied peak positions and