請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/61689
標題: | 運用藥物基因表現圖譜和疾病基因標記找尋疾病的藥物治療 Discovering Drug Treatment Using Drug Gene Expression Profiles and Disease Gene Signatures |
作者: | Wu-Lung Yang 楊伍隆 |
指導教授: | 趙坤茂(Kun-Mao Chao) |
共同指導教授: | 黃奇英(Chi-Ying Huang) |
關鍵字: | 藥物開發,基因微陣列,癌症用藥,機器學習,基因標記, drug development,microarray,cancer drugs,machine learning,gene signatures, |
出版年 : | 2013 |
學位: | 博士 |
摘要: | 藥物開發是一個費時且昂貴的過程。為了加快開發速度,我們運用老藥新用的概念。老藥新用是利用已核准上市的藥品於原適應症之外的用途。老藥新用的搜尋策略主要有兩種方法:(1)搜尋疾病和藥物間的關係,(2) 搜索藥物和藥物間的關係。
CMap是一個可篩選潛力藥物的基因表現圖譜資料庫。我們利用CMap搜尋疾病和藥物間的關係,透過疾病基因標記,可找尋和疾病基因標記成負相關的藥物基因表現,而推論此藥物能讓疾病恢復到正常狀態。此研究中,我們試用了幾種產生基因標記(gene signature)的方法,包括 T 檢定、集團(Clique)分析與頻率分析。我們分別在膽管癌(CAA)、鼻咽癌(NPC)和肝癌(HCC)等不同癌症進行研究。藥物預測的結果皆經由生物實驗驗証。我們認為, T 檢定為用於同質性來源資料的最佳方法,而集團分析則適用於異質性來源的資料。 此外,由於肝癌和鼻咽癌進行過大量的藥物實驗驗證, 因此我們以實驗結果結合支持向量機(Support Vector Machines)作機器學習。我們發現肝癌和鼻咽癌的化學敏感基因並進行第二輪的藥物預測。目前我們已建立可行性之藥物開發流程: 首先,運用CMap作初步的藥物預測,當藥物篩選結果充足時,再採用SVM進行更精確的效用預測。 另外,在搜尋藥物和藥物之間的關係時,我們建議採用SVM學習多種藥物在相同標靶的共同基因表現圖譜。這樣的預測模型將幫助了解藥物的作用並且幫助發現新藥物的標靶,進而促進藥物再利用的可能性。 最後,因為藥物和疾病都運用基因表現圖譜表現狀態,這能幫助我們整合藥物和疾病並推論出存在的相關性。為了擴充資料庫,在未來的研究裡,我們將整合NCBI的GEO基因微陣列資料庫和CMap,進而分析所有可能的藥物和疾病關係。 Drug discovery for diseases is a time-consuming and costly process. Drug repurposing, or the use of approved drugs outside of their original indications, will speed up the process. Two ways of drug repurposing strategies are proposed: one is to search for drug-disease association and the other is to search for drug-drug association. For drug-disease association, the Connectivity Map (CMap) is employed. CMap is an in-silico platform for screening gene expression patterns of diseases and drugs. A negative correlation of gene expression pattern between a disease and a drug will suggest that the drug might be able to revert the gene expression of the disease, thereby restoring the disease to a normal state. In this study, we employed several gene signature generation techniques, including t-test, clique analysis, and frequency analysis, to generate suitable signatures for Cholangiocarcinoma (CAA), Nasopharyngeal carcinoma (NPC), and Hepatocellular carcinoma (HCC). The predicted chemicals are biologically verified and the accuracies of these predictions are compared. We have shown that t-test is best for homogenous data source while clique analysis is best for heterogeneous data source. Furthermore, with the abundant drug screening results for HCC and NPC, by using Support Vector Machines, chemically sensitive genes for HCC and NPC are discovered and second-round of drug prediction is made. We have established a drug discovery pipeline to initially use CMap to predict first round of drugs and subsequently employ SVM for a more precise prediction while the number of positive samples are ample. For the drug-drug association, we propose to employ SVM to learn the drug-target pairs through the gene expression profiles. The learned models will help us uncover new drug-target relationships for existing chemicals and predict targets of new chemicals, thereby promoting possible drug-repurposing possibilities. The universal microarray proofing of chemical induced response and diseases have prompted the possibilities of integrating chemicals and diseases with the final goal of inferring the relationship among them. To broaden the database, we will utilize the NCBI GEO (Gene Expression Omnibus) microarray database in the future studies. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/61689 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 12.45 MB | 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。