Date of Award

May 2019

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Biomedical and Health Informatics

First Advisor

Susan McRoy

Committee Members

Hongfang Liu, Rohit J Kate, Jake Luo

Keywords

Biomedical literature, Drug repositioning, Natural Language Processing, Phenome-Wide Association Studies, Social media, Text data

Abstract

New drug development costs between 500 million and 2 billion dollars and takes 10-15 years, with a success rate of less than 10%. Drug repurposing (defined as discovering new indications for existing drugs) could play a significant role in drug development, especially considering the declining success rates of developing novel drugs. In the period 2007-2009, drug repurposing led to the launching of 30-40% of new drugs. Typically, new indications for existing medications are identified by accident. However, new technologies and a large number of available resources enable the development of systematic approaches to identify and validate drug-repurposing candidates with significantly lower cost. A variety of resources have been utilized to identify novel drug repurposing candidates such as biomedical literature, clinical notes, and genetic data. In this dissertation, we focused on using text data in identifying and prioritizing drug repositioning candidates and conducted five studies.

In the first study, we aimed to assess the feasibility of using patient reviews from social media to identify potential candidates for drug repurposing. We retrieved patient reviews of 180 medications from an online forum, WebMD. Using dictionary-based and machine learning approaches, we identified disease names in the reviews. Several publicly available resources were used to exclude comments containing known indications and adverse drug effects. After manually reviewing some of the remaining comments, we implemented a rule-based system to identify beneficial effects. The dictionary-based system and machine learning system identified 2178 and 6171 disease names respectively in 64,616 patient comments. We provided a list of 10 common patterns that patients used to report any beneficial effects or uses of medication. After manually reviewing the comments tagged by our rule-based system, we identified five potential drug repurposing candidates. To our knowledge, this was the first study to consider using social media data to identify drug-repurposing candidates. We found that even a rule-based system, with a limited number of rules, could identify beneficial effect mentions in the comments of patients. Our preliminary study shows that social media has the potential to be used in drug repurposing.

In the second study, we investigated the significance of extracting information from multiple sentences specifically in the context of drug-disease relation discovery. We used multiple resources such as Semantic Medline, a literature-based resource, and Medline search (for filtering spurious results) and inferred 8,772 potential drug-disease pairs. Our analysis revealed that 6,450 (73.5%) of the 8,772 potential drug-disease relations did not occur in a single sentence. Moreover, only 537 of the drug-disease pairs matched the curated gold standard in the Comparative Toxicogenomics Database (CTD), a trusted resource for drug-disease relations. Among the 537, nearly 75% (407) of the drug-disease pairs occur in multiple sentences. Our analysis revealed that the drug-disease pairs inferred from Semantic Medline or retrieved from CTD could be extracted from multiple sentences in the literature. This highlights the significance of the need for discourse-level analysis in extracting the relations from biomedical literature.

In the third and fourth study, we focused on prioritizing drug repositioning candidates extracted from biomedical literature which we refer to as Literature-Based Discovery (LBD). In the third study, we used drug-gene and gene-disease semantic predications extracted from Medline abstracts to generate a list of potential drug-disease pairs. We further ranked the generated pairs, by assigning scores based on the predicates that qualify drug-gene and gene-disease relationships. On comparing the top-ranked drug-disease pairs against the Comparative Toxicogenomics Database, we found that a significant percentage of top-ranked pairs appeared in CTD. Co-occurrence of these high-ranked pairs in Medline abstracts is then used to improve the rankings of the inferred drug-disease relations. Finally, manual evaluation of the top-ten pairs ranked by our approach revealed that nine of them have good potential for biological significance based on expert judgment.

In the fourth study, we proposed a method, utilizing information surrounding causal findings, to prioritize discoveries generated by LBD systems. We focused on discovering drug-disease relations, which have the potential to identify drug repositioning candidates or adverse drug reactions. Our LBD system used drug-gene and gene-disease semantic predication in SemMedDB as causal findings and Swanson’s ABC model to generate potential drug-disease relations. Using sentences, as a source of causal findings, our ranking method trained a binary classifier to classify generated drug-disease relations into desired classes. We trained and tested our classifier for three different purposes: a) drug repositioning b) adverse drug-event detection and c) drug-disease relation detection. The classifier obtained 0.78, 0.86, and 0.83 F-measures respectively for these tasks. The number of causal findings of each hypothesis, which were classified as positive by the classifier, is the main metric for ranking hypotheses in the proposed method. To evaluate the ranking method, we counted and compared the number of true relations in the top 100 pairs, ranked by our method and one of the previous methods. Out of 181 true relations in the test dataset, the proposed method ranked 20 of them in the top 100 relations while this number was 13 for the other method.

In the last study, we used biomedical literature and clinical trials in ranking potential drug repositioning candidates identified by Phenome-Wide Association Studies (PheWAS). Unlike previous approaches, in this study, we did not limit our method to LBD. First, we generated a list of potential drug repositioning candidates using PheWAS. We retrieved 212,851 gene-disease associations from PheWAS catalog and 14,169 gene-drug relationships from DrugBank. Following Swanson’s model, we generated 52,966 potential drug repositioning candidates. Then, we developed an information retrieval system to retrieve any evidence of those candidates co-occurring in the biomedical literature and clinical trials. We identified nearly 14,800 drug-disease pairs with some evidence of support. In addition, we identified more than 38,000 novel candidates for re-purposing, encompassing hundreds of different disease states and over 1,000 individual medications. We anticipate that these results will be highly useful for hypothesis generation in the field of drug repurposing.

Download

Share

COinS