Identification of Potential Adverse Drug Reactions using Random Walk on Network Models

Adverse drug event or adverse drug reactions (ADRs) are known to result from off-target pharmacological interactions and higher order effects. ADRs have been often ’discovered’ only after clinical events are reported and archived in databases such as FAERS. Off-target interactions and higher order effects are mostly reported at the pairwise and second order levels, respectively. Hence, many likely-ADRs are overlooked for a long time, leading to a multitude of potential morbi-mortality events with enormous public and individual health consequences. With the advent of high throughput omic technologies and large public databases, an enormous corpus of biochemical and phenotypic data is available to study ADRs. Detailed analyses of such data corpora may allow us to develop a deeper understanding of these relevant, higher order interactions, leading to a better assessment of potential ADRs. However, extracting useful information from such large data conglomerates is a challenging endeavor for the data sciences. Here, we applied methods from knowledge discovery in databases, network pharmacology, supervised and semi-supervised statistical learning, to tackle the problem of identifying potential adverse drug reactions. By analyzing a set of 500 drugs, including 67 systemic anti-infective agents, we identified over 27,000 potential adverse drug reactions based on activity on well-known biological function sets.