Accelerating Drug Development with Health-Consumer-Contributed Data and Heterogeneous Network Mining
出版項
2019
說明
1 online resource (147 pages)
文字
text
無媒介
computer
成冊
online resource
附註
Source: Dissertations Abstracts International, Volume: 81-08, Section: B
Advisor: Yang, Christopher C
Thesis (Ph.D.)--Drexel University, 2019
Includes bibliographical references
New drug development is a time-consuming, high-investment, and high-risk process. It usually takes more than ten years to bring a new medication to market and costs the pharmaceutical companies an average of $1.2 billion during the procedure. As a result, taking drugs that have been developed for one disease and "repositioning" them to another disease is becoming more and more common and an increasingly important strategy in both industry and academia. Current drug repositioning researches can be categorized as biological experimental methods and computational methods. The fast development of biomedical knowledge bases and computational capabilities has elevated a number of computational drug repositioning approaches, with less investment in time and cost compared with experimental approaches. Additionally, another important starting point for drug repositioning is off-label drug use. The detection of off-label drug uses in clinical practice can provide relevant hypotheses for drug repositioning and drug development, meanwhile, the findings in drug repositioning also provide possible off-label use opportunities. In this dissertation, we focus on both off-label drug use and drug repositioning to accelerate the drug development. With consideration of the insufficient utilization of unstructured biomedical data especially the contents contributed by health consumers, we employed health-consumer-contributed data for the topics. Specifically, for off-label drug use, we develop an automated method to detect off-label drug use from the heterogeneous healthcare network based on meta-path mining. In order to find better ways to represent the relationships between medical entities, we improve the method by introducing word embedding models to measure those associations on the basis of association rule mining. In order to deal with the sparsity and missing data problem in user-generated contents, we furtherly employ tensor decomposition techniques for detecting off-label drug-disease relationships. Experiment results show that the proposed approaches could identify off-label drug uses from health-consumer-contributed data effectively and accurately. Especially, when incorporating word embedding models and tensor decomposition, the models achieve better results. For drug repositioning, we propose a systematic method to identify repositioning drugs from health-consumer-contributed data by using the adverse drug reaction (ADR)-based repositioning strategy and heterogeneous network mining approaches. Based on the constructed heterogeneous healthcare network, we develop path-mining approaches to extract the significant associations between ADRs and diseases and then identify the novel associations between drugs and diseases for repositioning opportunity. To better understand the repositioning results and examine the computational repositioning approach, we conduct detailed literature review and case studies for a specific disease, Parkinson's Disease (PD). Additionally, in order to resolve the challenge of extracting ADR entities from health-consumer-contributed texts and improve the performance of drug repositioning detection, we utilize bidirectional long short-term memory (LSTM) models to recognize ADRs first and then use the expanded vocabulary to construct and mine the heterogeneous healthcare network. Experiment results show that ADRs are effective intermediaries to reveal drug-disease associations and the proposed approaches could suggest quite potential repositioning drug candidates
Electronic reproduction. Ann Arbor, Mich. : ProQuest, 2020
