Biomedical Engineering Reference
In-Depth Information
4.2
Linear Regression Methods ......................................................................................
182
4.3
Non-Linear Regression Methods..............................................................................
185
5
Conclusions........................................................................................................................
187
References................................................................................................................................
188
1 Introduction
Chapter 1 overviewed the state of routine and state of the art in bioprocess monitoring
and provided an indication of the amount and the complex nature of data monitored
routinely during bioprocess operation. Individual sensors measuring physical,
chemical and biological variables at various stages of the production process
typically yield extensive data sets with high sampling frequency. Sensor arrays, such
as electronic noses and tongues referred to in Chapter 1, and multianalyte analysers,
such as spectroscopic and fluorescence measurements, further increase the com-
plexity of the data structures collected and stored for 'post mortem' analysis of
process behaviour. A plethora of methods used to check the quality of the product
during various stages of processing, such as various chromatographic techniques and
ELISA assays, typically yield discrete data points with varying sampling frequencies
as well as accuracy. In addition a rapid expansion and improvements in methods of
genomic, trascriptomic, proteomic, metabolomic or environomic (Chap. 8) data
collection open up the possibilities of much better process understanding during the
development stage, as required by a quality by design (QbD) framework (Chaps. 4, 8
and 9). The latter methods result in data arrays with structures which are much more
complex and extensive than the data routinely monitored during process operation;
however these data also dramatically increase the potential benefits to be gained.
With such extensive data collection, the importance, and benefits of, effective
and robust data pre-processing and analysis become critical from both the
bioprocess development and monitoring and control points of view. This chapter
concentrates on data pre-processing, conditioning, reconciliation and multivariate
data analysis (MVDA) methods that enable advanced interpretation of process
measurement data for effective monitoring and control, whilst also being appli-
cable to process development. This chapter provides a basic overview of the
fundamental principles of both linear and non-linear methods most frequently used
for pattern recognition and regression building, with examples of application from
various stages of bioprocess operation. However, it is not possible to provide a
detailed description of each of the methods and all aspects of their applications.
Readers are referred to appropriate literature sources for more details on individual
techniques or relevant application examples.
It is important to stress that a number of data analysis methods described in this
chapter are well established and applied routinely in other scientific disciplines
(e.g. chemometric methods widely applied in chemistry) and industrial sectors,
and thus extensive resources, in terms of both literature and computing packages
Search WWH ::
Custom Search