Chemistry Reference
In-Depth Information
parallel and anti-parallel
-sheets, and mutations in nucleic acid. Thus it is important to assign the signals
while the spectral pattern is analyzed to obtain a reliable diagnostic result. The main focused spectral
differentiating information has been proposed for vibrations of methylene chains in membrane lipids,
hydrogen bonding of the phosphodiester groups of nucleic acids, glycogen content, hydrogen bonding of
C-OH groups in carbohydrates and proteins, symmetric phosphate stretching band of the phosphodiester
groups of nucleic acids and the methylene: methyl ratio.
β
21.4.2
The role of data processing
Although the signal assessment process has led to useful diagnostic results, it is inevitable to propose an
expandable medical method without proposing a reliable predictive model. This is the situation in which
chemometrics has an effective role, making the clinical approaches powerful. In this section we will briefly
review the main aspects of chemometrics as an emerging technology in all approaches to analytical chemistry
in order to define the requirements in clinical biodiagnostics. Chemometrics has always been a practically
oriented sub discipline of analytical chemistry for handling, interpreting and predicting chemical data. During
the past three decades, raw chemical data has increasingly been processed by computers. The interdisciplinary
field of chemometrics combines computing with applied mathematics, statistics and chemistry to extract
useful information from raw and large chemical data through univariate or multivariate methods. In case of
multivariate methods, chemometrics can be used in such areas as:
Outlier detection
●
Calibration
●
Classification and pattern recognition
●
Image Analysis
●
Of course the first three steps are the most common respective algorithms in chemometric based
data processing. Even though the image analysis diagnostic methods are also based on chemometric data
processing, they can be classified as pattern recognition.
21.4.2.1 Outlier detection
All results will be affected by error of input dataset. Ideal input data should be error free. However, analysts
can encounter outlier data. Outliers in multivariate data can severely impact the results of statistical analyses.
As referenced by Jouan-Rimbaud during the last decade, there are two types of chemical outliers:
1) The prediction samples contain the same components as the calibration samples, but in concentrations
that are outside the range in the calibration set, so that their prediction would lead to an extrapolation of
the model.
2)
The prediction samples contain components that were not present in the calibration samples (interference
causing species) [74].
Detection of data outliers and unusual data structures is one of the main tasks in the analysis of data. It is
very important to detect the outliers in a very precise way. Traditionally, despite the fact that data sets are
almost always multivariate, outliers are most frequently sought for each single variable in a given data set.
The standard diagnostics for outlier detection in the calibration phase of a multivariate calibration experiment
have been thoroughly discussed by Martens and Næs [75]. Accordingly, there are two approaches for outlier
detection. The first approach is to fit the data with least squares, construct regression diagnostics and then
remove the outliers. The second approach is to construct estimators that fit the majority of the data and
