S43 Assessing DDI Relevance Using Large Databases: From Social Media to Published Literature

Rion Brattig Correia , Indiana University, Bloomingotn, IN
Ian B. Wood , Indiana University, Bloomington, IN
Luis M. Rocha , Indiana University, Bloomington, IN
Biomedical literature mining can aid Drug-Drug Interaction (DDI) research by automatically identifying, classifying, and extracting evidence for large numbers of potential interactions from published scientific literature. Additionally, analysis of Electronic Health Records (EHR) and social media data via machine learning additionally provides individual- and population-level observation tools with the potential to speed translational research. Our most recent work has been on integrating and analyzing these different data sources in an effort to enhance public health monitoring and DDI research.

In what concerns literature mining, I will describe our work on automatic classification of PubMed abstracts and sentences for experimental evidence of DDI[1]. This is especially important given the sheer number of new articles that are being published on a daily basis. Our classifiers achieve high performance when deployed on labeled data and are able to uncover new evidence from unlabeled data, demonstrating that many papers in PubMed are missing MeSH terms to correctly identify them as containing evidence of DDI. I will also showcase our recent findings in correctly uncovering experimental evidence of DDI from PubMed publications, taking into account that different types of DDI evidence exist: in-vivo, in-vitro and clinical.

In regards to Social Media, we have recently demonstrated Instagram’s importance for public surveillance of DDI and adverse drug reactions (ADR)[2]. Our methodology is based on the longitudinal analysis of social media user timelines at different timescales. Graphs are built from the co-occurrence of terms from various biomedical dictionaries (drugs, symptoms, natural products, side-effects, and sentiment), and reveal relevant drug-drug and drug-symptom pairs, we well as clusters of terms and drugs associated with the complex pathology associated with depression. I will also talk about our recent findings in different digital cohorts using Twitter, Facebook and ChaCha timeline data.

Ultimately, our research aims at unifying these different sources of data in order to help patients, clinicians and researchers in the discovery and monitoring of DDI and ADR. I will end my talk showing a roadmap on how this can be achieved with examples from a city-wide analysis of EHR, which reveals the prevalence of prescribed DDI in southern Brazil.

References

[1] A. Kolchinsky, A. Lourenço, H. Wu, L. Li, and L.M. Rocha.[2015] "Extraction of Pharmacokinetic Evidence of Drug-drug Interactions from the literature." PLoS ONE 10(5): e0122199. doi:10.1371/journal.pone.0122199.

[2] R.B. Correia, L. Li, L.M. Rocha [2016]. "Monitoring potential drug interactions and reactions via network analysis of Instagram user timeliness". Pacific Symposium on Biocomputing. 21:492-503.