Biomedical Engineering Reference
In-Depth Information
utilising these methods, are available. However, bioprocessing, and in particular
the biopharmaceutical industry, can be said to lag significantly behind other
industrial sectors, such as the chemical industry and manufacturing, in terms of
routine application of these methods in large-scale processing. This is often
explained by the strict regulatory nature of the industry, although there are also
fundamental differences in the principles of application within each of these areas;
For example in the chemical analysis area where chemometric methods were
originally developed, there is usually a 'known' solution to the task, i.e. known
concentrations of analytes in question, whereas in biological applications, MVDA
techniques are often used as exploratory techniques to investigate what relation-
ships exist (or are evident from the data) between measured variables.
However, the successful implementation of the QbD and PAT initiatives,
introduced in this sector almost a decade ago, depends to a great extent on the
effective integration of the MVDA methods within the monitoring and control
framework. Thus, it is important that a wider understanding and application of
these methods within this industry sector is championed and illustrated on
successful, industrially relevant case studies. On the other hand, caution is advised
in the use of chemometric methods, for example by Brereton [
6
], who expresses a
concern at 'lots of people, often without a good mathematical or computational
background,
wanting
to
use'
them
quickly,
without
gaining
much
insight.
The following material is presented very much with this caveat in mind.
2 Data Pre-Processing
Data pre-processing and reconciliation is by no means a novel concept, and it is
widely applied in industrial facilities such as refineries or bulk chemical production
plants. As early as 1961 Kuehn and Davidson [
27
] described data reconciliation, and
alternative mathematical approaches have been explored since then with a multitude
of applications of data reconciliation in the field of engineering being suggested.
The importance of data pre-processing is universally acknowledged, and most
users of chemometric and MVDA methods would agree that the data preparation
steps take up most of the analysis time and can either 'make or break' the success
of any pattern recognition or model development efforts. One particular issue in
MVDA analysis is the removal of outliers, as highlighted by MØller et al. [
37
].
They argue that, since MVDA methods are usually based on a least-squares or
similar criterion, they are sensitive to outliers, which can lead to incorrect
conclusions. They describe a range of robust methods for outlier removal.
Further data pre-processing is then usually required, depending on a particular
application; For example in NIR spectra analysis, a range of specific data pre-
processing approaches has been developed (for more detail on pre-processing of
spectral data see Chap. 9). These include smoothing noisy data using a Savitzky-
Golay filter, baseline shift correction, multiplicative scatter correction (MSC) and
Search WWH ::
Custom Search