|
KMID : 0606320040200010047
|
|
Journal of Phamacetical Sciences Sookmyung Women s University
2004 Volume.20 No. 1 p.47 ~ p.53
|
|
|
Algorithm for finding the best regression models using NIR spectra for the analysis of biological samples
|
|
Hur Yun-Jeong
Cho Jung-Hwan
|
|
|
Abstract
|
|
|
|
An analytical method applicable to biological samples with minimum sampling and/or in noninvasive/nondestructive manners is highly recommended in diagnosis of disease status and in determination of course of treatment. A new noninvasive analytical method should be able to catch characteristics specific to target analytes of the biological samples. In other hands, near-infrared spectrophotometry is considered to be very useful for noninvasive analysis, in which eletromagnetic radiation permeable through the biological barrier is used. In the analysis of blood glucose, it is needed to extract minor signals of glucose from highly overlapped spectroscopic data. The conventional methods of principal component regression or any other factor-based regression analyses select the first several major factors with various criteria. The conventional methods are good for the major components, but they fail to give proper regression models for minor components such as blood glucose. To overcome this problem, all-possible combinations of principal factors are evaluated in terms of standard error of prediction(SEP) of glucose levels in the synthetic mixtures. This new algorithm is named APC-PCR(all-possible combination principal component regression). APC-PCR gave a good reduction of SEP in predicting glucose levels of synthetic mixtures. The maximum reduction of SEP was 21.6% with absorbance spectra, 25.8% with 1st derivative spectra and 25.3% with 2nd derivative spectra.
|
|
|
KEYWORD
|
|
|
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|