Page 309 - Data Science Algorithms in a Week

290 Jose M. Prieto

Antiviral Activities
Viruses are still a major, poorly addressed challenge in medicine. The prediction of
antiviral properties of chemical entities or the optimisation of current therapies to
enhance patient survival would be of great impact but the application of AI to this
conundrum has been less explored than in the case of antibacterials. Perhaps the most
pressing issue is the search for improved combination, antiretroviral drugs to suppress
HIV replication without inducing viral drug resistance. The choice of an alternative
regimen may be guided by a drug-resistance test. However, interpretation of resistance
from genotypic data poses a major challenge. Larder and co-workers (2007) trained
ANNs with genotype, baseline viral load and time to follow-up viral load, baseline CD4+
T-cell counts and treatment history variables. These models performed at low-
intermediate level, explaining 40-61% of the variance. The authors concluded that this
was still a step forward and that these data indicate that ANN models can be quite
accurate predictors of virological response to HIV therapy even for patients from
unfamiliar clinics.
We recently tried to model the activity of essential oils on herpes viruses (types 1 and
2) by both MLR and ANNs (Tanir & Prieto, unpublished results). Our results could not
find a clear subset of chemicals with activity, but rather the best results were given by
datasets representing all major components. This highlights that viruses are a much
harder problem to model and more work must be done towards solving it.

Prediction of Pharmacological/Toxicological Effects and Disease Biomarkers

The prediction of pharmacological or toxicological effects should ideally involve
whole living organisms or at least living tissues. However, the current approach is the use
of culture mammal cells, favouring single proteins as targets. Therefore, predicting these
effects is clearly more complex than the prediction of purely chemical reactions (such as
antioxidant activities) or antimicrobial ones (bacteria, fungi, viruses).
Inflammation is the response of a living tissue to an injury. Therefore, is
fundamentally a multifactorial process which may pose extreme complexity on its
modeling. An approximation to the problem is to target the inhibition of key enzymes
responsible for the onset and maintenance of such process such as cyclooxygenases and
lipoxygenases. Nonsteroidal anti-inflammatory drugs inhibiting either of those targets are
the most used anti-inflammatory medicines in the world. Dual inhibitors of
cyclooxygenase-1 and 5-lipoxygenase are proposed as a new class of anti-inflammatory
drugs with high efficacy and low side effects. In a recent work, Chagas-Paula and co-
workers (2015) selected c.a. 60 plant leaf extracts from Asteraceae species with known in
vitro dual inhibition of cyclooxygenase-1 and 5-lipoxygenase and analyzed them by
HPLC-MS-MS analysis. Chromatographic peaks of the extracts were correlated to their