Page 308 - Data Science Algorithms in a Week

Artificial Intelligence for the Modeling and Prediction ... 289

(2000). The challenge of the complexity of the countless chemical interactions between
dozens of EOs components and the microbes is virtually impossible to address in the
laboratory, but it may be solved using computational models such as artificial neural
networks (ANNs). In addition, ANNs are theoretically able to consider synergies and
antagonisms between inputs. There is a consistent body of data on many crude essential
oils being more active than their separated fractions or components, report on synergies.
In some cases synergistic activity between two or three components could be
experimentally demonstrated (Didry et al., 1993; Pei et al., 2009), but to do so with
dozens of chemicals is beyond reach. In fact, ANNs are algorithms which has the
capacity of approximating an output value based on input data without any previous
knowledge of the model and regardless the complexity of its mechanisms, in this case the
relationship between the antioxidant capacity of a given essential oil (input data) and its
chemical composition (parameters affecting the assay). The enormous amount of
information produced on the antimicrobial activity of essential oils provides a rich field
for data-mining, and it is conceivable to apply suitable computational techniques to
predict the activity of any essential oil by just knowing its chemical composition.
Our results reflect both the variability in the susceptibility of different
microorganisms to the same essential oil, but more importantly point towards some
general trends. The antimicrobial effects of essential oils upon S. aureus and C.
perfringens (Gram +) were accurately modelled by our ANNs, thus meaning a clear
relationship between the chemistry of EOs and their susceptibility, perhaps suggesting a
more additive, physical -rather than pharmacological- mechanism of action. This also
opens the prospect for further studies to ascertain the best set of volatile components
providing optimum antimicrobial activity against these two pathogens and/or Gram + in
general. On the other hand, the lower accuracy of the predictions against E. coli (Gram -)
and C. albicans (yeast) may suggest more complex pharmacological actions of the
chemicals. In this case the activity may be pinned down to one or few active principles
acting individually or in synergies.
Ozturk et al. (2012) studied the effects of some plant hydrosols obtained from bay
leaf, black cumin, rosemary, sage, and thyme in reducing Listeria monocytogenes on the
surface of fresh-cut apple cubes. In addition to antibacterial measurements, the abilities of
Adaptive neuro-fuzzy inference system (ANFIS), artificial neural network (ANN), and
multiple linear regression (MLR) models were compared with respect to estimation of the
survival of the pathogen. The results indicated that the ANFIS model performed the best
for estimating the effects of the plant hydrosols on L. monocytogenes counts. The ANN
model was also effective but the MLR model was found to be poor at predicting
microbial numbers. This further proofs the superiority of AI over Multivariate statistical
methods in modeling complex bioactivities of chemically complex products.