Page 303 - Data Science Algorithms in a Week

284 Jose M. Prieto

On the one hand, application of ANNs to food technology, for example control of
bread making, extrusion and fermentation processes (Batchelor, 1993; Eerikanen &
Linko, 1995; Latrille, Corrieu, & Thibault, 1993; Ruan, Almaer, & Zhang, 1995) are
feasible and accurate, easy to implement and will result in noticeable advantages and
savings to the manufacturer. On the other hand, prediction of functionality (antioxidant,
antimicrobial activities for example) is not so much explored, perhaps given the
complexity of the experimental design associated with those, that we will discuss in detail
later, and the less obvious advantages for the manufacturer.
The potential applications of ANN methodology in the pharmaceutical sciences
range from interpretation of analytical data, drug and dosage form design through
biopharmacy to clinical pharmacy. This sector focuses more on the use of ANNs to
predict extraction procedures (similarly to the food sector), pharmacokinetic and
toxicological parameters. These three aspects are usually non-linear thus in need of AI
tools, that can recognize patterns from data and estimate non-linear relationships. Their
growing utility is now reaching several important pharmaceutical areas, including:

 Quantitative Structure Activity Relationship (QSAR) and molecular modeling
(Kovesdi et al., 1999; Jalali-Heravi & Parastar, 2000)
 Toxicological values of organic compounds based on their structure and
mutagenicity (Jezierska, Vračko, & Basak, 2004).
 Pharmacological activities (Chen et al., 2011)
 Modeling of drug solubility (Huuskonen, Salo, & Taskinen, 1998) and other
pharmacokinetic parameters (Ma et al., 2014)
 Response surface modeling in instrumental (chromatography) to predict the
retention as a function of changes in mobile phase pH and composition analysis
optimization (Agatonovic-Kustrin & Loescher, 2013)
 Optimization of formulations in pharmaceutical product development (Parojcić et
al., 2007)

Most of the above problems are solved for the case of single (natural or synthetic)
drugs. However, the urgency of applying ANN based approaches is best perceived to the
clinical rationalisation and exploitation of herbal medicines. Herbal medicines contain at
least one plant based active ingredient which in turn contains dozens to hundreds of
components (phytochemicals). To start with, little is known about which phytochemical/s
is/are responsible for the putative properties of the herbal ingredient. Chagas-Paula et al.
(2015) successfully applied ANNs to predict the effect of Asteraceae species which are
traditionally used in Europe as anti-inflammatory remedies (for details see “Prediction of
the anti-inflammatory activities” below). When multiple herbal ingredients (10-20) are
used, such as in Traditional Chinese Medicine, the exact role of each drug may be only
possible to understand if the myriad of influencing factors are harnessed by AI means.