Page 304 - Data Science Algorithms in a Week

Artificial Intelligence for the Modeling and Prediction ... 285

(Han, Zhang, Zhou, & Jiang, 2014) taking advantage of the fact that ANNs require no
knowledge of the internal mechanism of the processes to be modelled.
Similar to pharmacotoxicology, pathology is a complex field in which modern High-
throughput biological technology can simultaneously assess the levels of expression of
tens of thousands of putative biomarkers in pathological conditions such as tumors, but
handling this complexity into meaningful classification to support clinical decisions
depends on linear or non-linear discriminant functions that are too complex for classical
statistical tools. ANNs can solve this issue and to provide more reliable cancer
classification by their ability to learn how to recognize patterns (Wang, Wong, Zhu, &
Yip, 2009)

Prediction of Antioxidant Properties

Antioxidant capacity is nowadays accepted as a criterion of food quality and to
monitor the impact of food processing in the nutraceutical value of food products
(Shahidi, 2000). In experimental pharmacology antioxidant properties are also the object
of intense research as they have been shown to influence and resolve many pathological
processes (Young & Woodside, 2001), but so far, the complexity and sometimes
contradictory effects of antioxidants hamper their implementation into therapeutic
approaches (Mendelsohn & Larrick, 2014). Therefore, developing ANNs able to predict
antioxidant values of natural products may become an important tool for the food
industry as they could avoid implementing any experimental procedure within their
premises. The antioxidant properties of natural products have been on the centre of
intensive research for their potential use as preservatives, supplements, cosmeceuticals or
nutraceuticals by the food and cosmetics industry. Literally hundreds of works reporting
both on the composition and antioxidant properties of natural products have been written
during the last decade. However, this kind of work is under an increasing criticism as the
inherent intra-specific variability of their composition -depending on the location,
altitude, meteorology, type of soil and many other factors- make this kind of work
virtually irreproducible.
To our knowledge, the first report showing the possibility of applying ANNs to
predict the antioxidant capacity of natural products was presented by Buciński, Zieliński,
& Kozłowska, (2004). The authors chose to use the amount of total phenolics and other
secondary metabolites present in cruciferous sprouts as input data. Despite the popularity
of this topic in natural products chemistry no further attempts to use an ANN for the
prediction of the antioxidant capacity of natural products was done until our pioneering
work to predict the antioxidant activity of essential oils in two widely used in vitro
models of antiradical and antioxidant activity, namely 2,2-diphenyl-1-picrylhydrazyl
(DPPH) free radical scavenging activity and linoleic acid oxidation. We could predict the