請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/51240
標題: | 以排序學習方法整合多元資訊實現藥物重新定位 Computational Drug Repositioning: A Learning to Rank Approach with Multiple Data Sources |
作者: | Yu-Ting Chang 張宇婷 |
指導教授: | 魏志平(Chih-Ping Wei) |
關鍵字: | 舊藥新用,醫學文獻探勘,監督式學習,排序學習,文檔列表方法, drug repositioning,biomedical literature mining,supervised learning,learning to rank,list-wise approach, |
出版年 : | 2015 |
學位: | 碩士 |
摘要: | 藥物開發過程繁複,其成本高且成功率低。根據美國食品藥品管理局(FDA)規定,新藥物需通過動物、人體實驗、委員會審核等七大流程才可以在市場中販售。然而,只要其中一個流程無法通過,此開發工程就前功盡棄,開發藥廠將承受所有的損失。因此,藥物開發人員開始引入「舊藥新用」的開發方式,從既有藥物中,在其原本設計標的外,尋找新適應症。若能從將既有藥物重新定位、找到新用途,開發藥廠就可因此減少開發的時間和成本。
Swanson (1986)最先提出以醫學文獻探勘實現藥物重新定位的技術,而Chen (2013)基於Swanson的ABC模型,進行改良並納入醫藥知識庫,提出Path模型。本研究結合了醫學文獻及醫藥知識庫,使用ABC和Path模型以不同排序演算法產生的藥物對適應症之排序結果作為特徵,以排序學習(Learning to Rank)中的文檔列表方法(List-wise Approach)構建藥物重新定位系統–DRLTR。 我們以各個特徵及其他排序學習方法對DRLTR進行比較性實驗,結果顯示DRLTR都有顯著的進步,表示我們所提出的方法可以更有效地提供潛在的藥物及疾病關係給研究者,以協助研究人員進行藥物的重新定位。 Drug development is costly and time consuming. According to the regulation of United States Food and Drug Administration (FDA), drug development consists of eight stages, including animal and human test, application review, etc. However, once one of the stages fails, the investment on candidate drug seldom returns. Therefore, drug developers and researchers start to explore the drug repositioning approach for drug development. If they can successfully find novel indications for an existing drug, the pharmaceutical company could save a lot of time and money. Swanson (1986) first proposed a drug repositioning technique by mining biomedical literature database. Chen (2013) refined Swanson’s ABC Model and considered ontologies into his Path Model. In this study, we combine both literature- and ontology-based databases, apply the ABC and Path Model with different ranking algorithms to generate several ranking scores as our features, and then propose a Drug Repositioning Learning-to-Rank System (DRLTR) that employs the list-wise learning-to-rank approach for effectively determining the final ranking. In our systematic evaluation experiments, we take each type of features and other learning to rank methods to compare with our proposed DRLTR system. As our experiment results show, DRLTR can effectively predict potential drug-disease relationships for drug developers and help them on the research of drug repositioning. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/51240 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊管理學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-104-1.pdf 目前未授權公開取用 | 756.38 kB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。