332
intensities and concluded that similarity functions, such as correlation coefficient
and Euclidean cosine, together with their first derivative variations should be used
in windows where their values are more pronounced, which means that the differences in spectra will be highlighted. In order to decide what the values of the above
parameters should be, one must devise an algorithm for a specific case, implement
it on a computer, and select the best values by trial and error method.
Spectral correlation analysis can be considered as the preprocessing step for the
identification of substances in the mixture of several components taken from the
collection of substances that we are concerned with, by mixture spectrum and template spectra of substances. If we assume the linear model for a mixture spectrum,
i.e., it is a linear combination of template spectra, the identification is an optimization process where we look for optimum component coefficients that minimize or
maximize a comparison measure. In this situation it is evident that the more similar
the spectra, the more difficult is their identification. thus, it is important to perform
this correlation analysis for different correlation measures and different parameters
mentioned above in order to select the most appropriate ones. the values of the
measures for given parameters and all pairs of template spectra can help in the
determination of estimates from below (above), i.e., minimum (maximum) values
for which we can accept a potential component as part of a mixture. the estimation from below (above) is related to the similarity (dissimilarity) measure. Let us
remind here that the Pearson correlation coefficient is a similarity measure, whereas
the Euclidean distance or the Euclidean cosine are dissimilarity measures.
thus, if the value of a similarity function for a template and mixture spectra is
substantially greater than its maximum value for all pairs of template spectra from
the database, then we have a strong indication for the template to be part of the
mixture. We can relax this condition by taking the maximum value of similarity
function for this particular template and the rest of the spectra from the database.
For the template spectrum that can be easily differentiated from the rest (e.g., Phenylalanine, vide infra) the condition is considerably weaker.
12.3 Results
We performed spectral similarity investigation for 20 most important amino acids
coded by living organisms. they are listed in table 12.1. As templates, we used the
spectra recorded separately for each amino acid, using the laser line of 632.8 nm.
the experimental spectra are presented in Figs. 12.1 and 12.2. Changing the laser
wavelength does not have much influence on the spectra with respect to both, band
positions and intensities. the amino acids containing the phenyl substituent exhibit
absorption at the uv line; because of fluorescence, it was not possible to obtain the
spectra for tryptophan and tyrosine for 325 nm excitation. the spectra of phenylalanine could be registered, but they were strongly overlapping with the emission.
Comparison of Raman spectra of a large group of proteinogenic amino acids in
the solid phase has been, until now, presented in two papers. A work by Zhu et al.
T. Roliński et al.
intensities and concluded that similarity functions, such as correlation coefficient
and Euclidean cosine, together with their first derivative variations should be used
in windows where their values are more pronounced, which means that the differences in spectra will be highlighted. In order to decide what the values of the above
parameters should be, one must devise an algorithm for a specific case, implement
it on a computer, and select the best values by trial and error method.
Spectral correlation analysis can be considered as the preprocessing step for the
identification of substances in the mixture of several components taken from the
collection of substances that we are concerned with, by mixture spectrum and template spectra of substances. If we assume the linear model for a mixture spectrum,
i.e., it is a linear combination of template spectra, the identification is an optimization process where we look for optimum component coefficients that minimize or
maximize a comparison measure. In this situation it is evident that the more similar
the spectra, the more difficult is their identification. thus, it is important to perform
this correlation analysis for different correlation measures and different parameters
mentioned above in order to select the most appropriate ones. the values of the
measures for given parameters and all pairs of template spectra can help in the
determination of estimates from below (above), i.e., minimum (maximum) values
for which we can accept a potential component as part of a mixture. the estimation from below (above) is related to the similarity (dissimilarity) measure. Let us
remind here that the Pearson correlation coefficient is a similarity measure, whereas
the Euclidean distance or the Euclidean cosine are dissimilarity measures.
thus, if the value of a similarity function for a template and mixture spectra is
substantially greater than its maximum value for all pairs of template spectra from
the database, then we have a strong indication for the template to be part of the
mixture. We can relax this condition by taking the maximum value of similarity
function for this particular template and the rest of the spectra from the database.
For the template spectrum that can be easily differentiated from the rest (e.g., Phenylalanine, vide infra) the condition is considerably weaker.
12.3 Results
We performed spectral similarity investigation for 20 most important amino acids
coded by living organisms. they are listed in table 12.1. As templates, we used the
spectra recorded separately for each amino acid, using the laser line of 632.8 nm.
the experimental spectra are presented in Figs. 12.1 and 12.2. Changing the laser
wavelength does not have much influence on the spectra with respect to both, band
positions and intensities. the amino acids containing the phenyl substituent exhibit
absorption at the uv line; because of fluorescence, it was not possible to obtain the
spectra for tryptophan and tyrosine for 325 nm excitation. the spectra of phenylalanine could be registered, but they were strongly overlapping with the emission.
Comparison of Raman spectra of a large group of proteinogenic amino acids in
the solid phase has been, until now, presented in two papers. A work by Zhu et al.
T. Roliński et al.