DrugClust: A machine learning approach for drugs side effects prediction

Abstract

Background: Identification of underlying mechanisms behind drugs side effects is of extreme interest and importance in drugs discovery today. Therefore machine learning methodology, linking such different multi features aspects and able to make predictions, are crucial for understanding side effects.

Methods: In this paper we present DrugClust, a machine learning algorithm for drugs side effects prediction. DrugClust pipeline works as follows: first drugs are clustered with respect to their features and then side effects predictions are made, according to Bayesian scores. Biological validation of resulting clusters can be done via enrichment analysis, another functionality implemented in the methodology. This last tool is of extreme interest for drug discovery, given that it can be used as a validation of the clusters obtained, as well as for the study of new possible interactions between certain side effects and nontargeted pathways.

Results: Results were evaluated on a 5-folds cross validations procedure, and extensive comparisons were made with available datasets in the field: Zhang et al. (2015), Liu et al. (2012) and Mizutani et al. (2012). Results are promising and show better performances in most of the cases with respect to the available literature.

Availability: DrugClust is an R package freely available at: https://cran.r-project.org/web/packages/DrugClust/index.html.

Keywords: Drugs side effects; Machine learning; R package DrugClust.