Drug Side-Effect Prediction Based on the Integration of Chemical and Biological Spaces

Drug side-effects, or adverse drug reactions, have become
a major
public health concern and remain one of the main causes of drug failure
and of drug withdrawal once they have reached the market. Therefore,
the identification of potential severe side-effects is a challenging
issue. In this paper, we develop a new method to predict potential
side-effect profiles of drug candidate molecules based on their chemical
structures and target protein information on a large scale. We propose
several extensions of kernel regression model for multiple responses
to deal with heterogeneous data sources. The originality lies in the
integration of the chemical space of drug chemical structures and
the biological space of drug target proteins in a unified framework.
As a result, we demonstrate the usefulness of the proposed method
on the simultaneous prediction of 969 side-effects for approved drugs
from their chemical substructure and target protein profiles and show
that the prediction accuracy consistently improves owing to the proposed
regression model and integration of chemical and biological information.
We also conduct a comprehensive side-effect prediction for uncharacterized
drug molecules stored in DrugBank and confirm interesting predictions
using independent information sources. The proposed method is expected
to be useful at many stages of the drug development process.