|
 
|
About
Chemometrics Most human perceptions are a result of phenomena where multiple compounds together produce a physical (haze, foam or color) or direct sensory phenomenon (olfaction or taste). As a result multivariate methods are needed to successfully study and develop understanding of these phenomena. Chemometrics is the application of statistical and mathematical methods as well as the principles of measurement science and experimental design to efficiently extract maximum useful information from chemical data. This type of analysis can also be applied to sensory and biological measurements and typically is applied when multiple measurements are made on a set of samples. Chemometrics can help elucidate the nature of relationships between product composition and sensory properties and between instrumental measurements and sensory properties. Statistical techniques commonly applied include exploratory data analysis, pattern recognition, and empirical modeling. Exploratory data analysis (EDA) is often used to simplify and gain better understanding of large, complicated data sets. Two main procedures used for EDA are principal components analysis and cluster analysis. EDA can also be used to determine how many fundamental properties are represented in a data set and the extent to which measurements are redundant. This can be used to intelligently reduce analysis work while sacrificing little, if any, information. Pattern recognition (PARC) is useful for discerning classification rules that characterize groups of samples. For example, PARC can be used to identify the cultivar or growing area of a raw material. The brand or production plant in which a product was made can be identified from its pattern of analytical results. PARC can also be used to identify microorganisms, such as contaminating bacteria. Advanced PARC procedures can detect adulteration or be used for multivariate QA/QC, where determination of whether samples are inside or outside a multidimensional envelope of good quality is desired. Modeling
is the collection
of data in order
to develop a
formula that
describes the
behavior of
a system. Empirical
modeling has
many brewing
applications
including development
and optimization
of analytical
methods, discerning
relationships
between product
composition
and sensory
properties,
developing knowledge
of relationships
between molecular
structure and
biological or
physical properties
of compounds,
optimizing process
performance
and/or cost,
and developing
control algorithms
for unit operations
or processes. Questions? © Copyright 2002 by the American Society of Brewing Chemists. All rights reserved. |
