Predicting Adverse Drug Effects from Literature- and Database-Mined Assertions

Abstract

Introduction: Given that adverse drug effects (ADEs) have led to post-market patient harm and subsequent drug withdrawal, failure of candidate agents in the drug development process, and other negative outcomes, it is essential to attempt to forecast ADEs and other relevant drug-target-effect relationships as early as possible. Current pharmacologic data sources, providing multiple complementary perspectives on the drug-target-effect paradigm, can be integrated to facilitate the inference of relationships between these entities.

Objective: This study aims to identify both existing and unknown relationships between chemicals (C), protein targets (T), and ADEs (E) based on evidence in the literature.

Materials and methods: Cheminformatics and data mining approaches were employed to integrate and analyze publicly available clinical pharmacology data and literature assertions interrelating drugs, targets, and ADEs. Based on these assertions, a C-T-E relationship knowledge base was developed. Known pairwise relationships between chemicals, targets, and ADEs were collected from several pharmacological and biomedical data sources. These relationships were curated and integrated according to Swanson's paradigm to form C-T-E triangles. Missing C-E edges were then inferred as C-E relationships.

Results: Unreported associations between drugs, targets, and ADEs were inferred, and inferences were prioritized as testable hypotheses. Several C-E inferences, including testosterone → myocardial infarction, were identified using inferences based on the literature sources published prior to confirmatory case reports. Timestamping approaches confirmed the predictive ability of this inference strategy on a larger scale.

Conclusions: The presented workflow, based on free-access databases and an association-based inference scheme, provided novel C-E relationships that have been validated post hoc in case reports. With refinement of prioritization schemes for the generated C-E inferences, this workflow may provide an effective computational method for the early detection of potential drug candidate ADEs that can be followed by targeted experimental investigations.

Fig. 1

Workflow for selecting study chemicals from the NPC database. NPC: NCGC Pharmaceutical Collection; MW: molecular weight

Fig. 2

Time-split validation workflow. CTE: Chemical-Target-Effect; Y: year, C-E: Chemical-Effect; C-T: Chemical-Target; T-E: Target-Effect

Fig. 3

Network visualization. (a) Network visualization of interconnected chemicals (blue), targets (green), and ADEs (red) interconnections (b) Connectivity clusters in each domain (c) A subnetwork for the paroxetine-thrombocytopenic purpura inference

Fig. 4

Comparison of known vs. inferred C-E edges (coverage of known C→E edges = 94%)

Fig. 5

Substudy 2 scoring system performance, derived from raw data. (a) ROC curve, (b) ROC enrichment curve, and (c) Precision-Recall curve

Fig. 6

Time-split validation results. (a) Post-cutoff year coverage rates from time-split validation. C→E inferences that could be made with C→T and T→E edges published prior to certain years (1987–2007 in 5-year intervals), and their coverage of the known C→E edges published after that given year. The total number of known C→E edges established after a given cutoff year is listed above each bar. (b) Pre-cutoff year validation rates from time-split validation. The percentage of C→E inferences that could be made prior to a given cutoff year that were later validated. The total number of inferred C→E edges that could be established prior a given cutoff year is listed above each bar