There is growing interest in investigating the biochemical pathways involved in cellular responses to drugs. Here we propose new methods to explore the relationships between drugs, biochemical pathways and adverse drug reactions (ADRs) at a large scale. Using sparse canonical correlation analysis of 832 drugs characterized by 173 pathways and 1385 ADRs profiles, we identified 30 highly correlated sets of drugs, pathways and ADRs. This included known and potentially novel associations. To evaluate the predictive performance of our method, the extracted correlated components were used to predict known ADR profiles from drug pathway profiles. A relatively high prediction performance (AUC: 0.894) was achieved. To further investigate their association, we developed a network-based approach to extracting potentially significant modules of pathway-ADR associations. Five statistically significant modules were extracted. We found that most of the nodes contained in the modules are either pathways linked to a very limited number of drugs or rare ADRs. The work provides a foundation for future investigations of ADRs in the context of biochemical pathways under different clinical conditions. Our method and resulting datasets will aid in: a) the systematic prediction of ADRs, and b) the characterization of novel mechanisms of action for existing drugs. This merits additional research to further assess its potential in improving personalized drug safety monitoring, as well as for the repositioning of drugs in the longer term.
- Biological information theory
- Adverse drug reactions
- biological pathways
- sparse canonical correlation analysis