Neural computing in pharmaceutical products and process development - Computer-Aided Applications in Pharmaceutical Technology - page 111

clusters in a single step (Jain et al., 1999). Data clustering is often
accompanied by the dimensionality reduction and simplifi cation of the
system. Dimensionality reduction is usually based upon feature extraction
or feature selection within the data set.
One of the most often used neural networks, which is based upon
unsupervised learning, is the self-organizing map (SOM), described in
more detail in Section 5.5.

Limitations of the ANN

Note that ANNs cannot decipher input-output relationships in a
mechanistic sense, therefore the true nature of the infl uence of independent
variables on dependent ones remains unknown. Furthermore, a considerable
amount of data is often needed to build a reliable ANN model. Many
important steps in building and testing of the ANN are prone to errors and
diffi culties in optimal determination. In addition, each software tool used is
different and has its own distinctive properties. It is always recommended
to fi rst thoroughly investigate the principles upon which the neural
networks models are built and tested in the software used.

5.1.3 Examples

ANNs were applied in many cases of pharmaceutical product development
(Hussain et al., 1991; Takahara et al., 1997; Takayama et al., 1999a,b,
2000; Ibric et al., 2002; Plumb et al., 2002; Vaithiyalingam and Khan,
2002; Kachrimanis et al., 2003; Lim et al., 2003).
In pharmaceutical modeling and/or optimization studies, experiments
are usually organized according to an experimental design. These
experiments can also be studied using ANNs, in order to compare
different approaches for data analysis. Also, one of the main advantages
of ANNs, in comparison to traditional modeling and statistical analysis
techniques, is that the data used for artifi cial network training do not
need to be organized according to formal experimental design, and more
informal designs can be used. This is especially useful when organization
of experiments according to experimental design produces practical
problems due to the impossibility to conduct certain experimental runs.
Numerous examples of the ANN application in pharmaceutical
research, product development, and process optimization are available in
the literature (Table 5.2). Some of them are presented here, giving a short
description of the techniques used.

￿

￿ ￿