Neural computing in pharmaceutical products and process development - Computer-Aided Applications in Pharmaceutical Technology - page 154

as a reduced danger of overfi tting or overtraining (Manallack and
Livingstone, 1999).
Data clustering ability of unsupervised learning algorithms is diffi cult
to test. One of the proposed methods involves resampling of the data
once the initial clustering is complete. Different random subsets of data
are re-clustered and compared to original clustering results, in order to
check the robustness of the SOMs (Levine and Domany, 2001).

5.5.3 Examples

SOMs are used in pattern recognition, robotics, engineering, astronomy,
chemistry, medicine, biology, fi nance, process analysis, machine
perception, control, and communication. There are also numerous
examples of SOM usage in the fi eld of pharmaceutical products and
process development and optimization. SOMs were used in QSAR
studies (Manallack and Livingstone, 1999; Guha et al., 2004), new drug
design (Manallack and Livingstone, 1999; Shneider and Nettekoven,
2003; Kaiser et al., 2007), pharmacophore mapping (Polanski, 2003),
modeling studies of human oral bioavailability of drugs (Wang et al.,
2008), investigation of causal relationships in pharmaceutical
formulations (Yasuda et al., 2010), visualization of particle size and
shape distributions (Laitinen et al., 2002), and classifi cation of powder
fl owability and prediction of tablet weight variation (Antikainen et al.,
2000), etc.
Several examples of the SOM usage are presented here:

￿

￿ ￿

Example 1

SOMs were used to analyze relationships between formulation
factors, latent variables, and release properties of diltiazem
hydrochloride (DTZ) hydrophilic matrix tablets (Kikuchi et al., 2011).
Feature maps were used to analyze both global and local correlations
between variables (Figure 5.16). All plots depict the distribution of
each variable:

■ formulation factors

(a) the amount of dextran sulfate (DS);