gms | German Medical Science

Background: The causal analysis of potential adverse drug reactions (ADRs) is verychallenging due to their mechanistic complexity (multiple genes, genetic variations, proteins, and their interaction are relevant) [1]. To investigate the cause of ADRs not only the active agent but also contributory factors in the causal chain have to be studied [2]. Integrative omics help to un derstand disease pathogeneses and drug toxicity at molecular level and to apply predictive models in clinical research and practice [3].

Objective: The problem of identifying effects that are causal to the drug use is the need of qualitative data rather than disease- side effect associations. Adverse reactions include many factors that interact in the causal mechanism. These factors are essential for the effect, but not causal themselves (e.g. genetic susceptibility due to CYP p450 defects). They are categorized as clinical factors (dose responsiveness, time course, comorbidities, etc.) and biological factors (molecular interaction like inhibition of therapeutic targets or key proteins in physiological pathways) [2], [1]. The goal of the tool ADRrisk is to allow the exploration of associated entities on molecular level and thereby develop hypotheses between links and identify risk-modifying factors.

Methods: The associations between therapeutic indications and ADRs are measurable changes of patient phenotype that can be used to study the relationship between disease and drug toxicity [4]. Therefore, we integrated and unified individual case data obtained from pharmacovigilance databases (FDA FAERS, Canadian CVARD and Yellow Card Scheme from the UK [5], [6], [7]) to find significant associations between indication and ADR. We implemented the web application RiskADR to visualize the obtained associations in a network and display the connected molecular information for further analysis. Molecular links are queried from a heterogeneous network integrated in a Neo4j database, that contains the relevant biological entities (e.g. gene, disease, pathway, protein, variant) involved in action of a drug using established databases like Uniprot, Gene Ontology, and Reactome. The database ADReCS-Target, for instance, summarizes information on ADR, gene variant and protein associations obtained from curated literature mining. Using the graph database, disease-ADR associations can be analyzed by determining the shortest path in the network between the node entities. RiskADR is a shinyR web application [8] that incorporates the vis.js javascript library for network visualization and RNeo4j connector.

Results and Discussion: Causation of adverse drug reactions is a complex issue that requires indepth knowledge of clinical aspects as well as knowledge of the underlying molecular processes. The result set contains 174,485 disease-ADR associations obtained by logistic regression models controlling for number ofdrugs, sex and age group. For each association FDR-adjusted p-values and association measures are displayed to filter significant pairs of interest. Applications like RiskADR contribute to systems medicine and helpto mechanistically analyze drug-ADR pairs and are a promising solution to the complex problem of understanding mechanisms of toxicity.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.