View : 608 Download: 0

Full metadata record

DC Field Value Language
dc.contributor.advisor최선-
dc.contributor.author이선경-
dc.creator이선경-
dc.date.accessioned2016-08-26T11:08:36Z-
dc.date.available2016-08-26T11:08:36Z-
dc.date.issued2010-
dc.identifier.otherOAK-000000057184-
dc.identifier.urihttps://dspace.ewha.ac.kr/handle/2015.oak/203684-
dc.identifier.urihttp://dcollection.ewha.ac.kr/jsp/common/DcLoOrgPer.jsp?sItemId=000000057184-
dc.description.abstractAdenosine receptors (ARs) belong to the rhodopsin family of G-protein coupled receptors, and they can be classified into four subtypes A1, A2A, A2B, and A3. Among them, the most recently found subtype A3 AR is a promising target for the clinical drugs. A3 AR agonists are expected to reduce myocardial ischemia and reperfusion injury, and antagonists are being investigated as anti-asthma, anti-glaucoma and anti-inflammatory agents. In this study, two ligand-based approaches were applied to obtain further idea of designing agonists and antagonists. Classification models were built to distinguish the antagonists from the agonists by employing three machine learning methods, Laplacian-modified naive Bayesian, recursive partitioning, and support vector machine, which are widely used as classification tools in the early stage of drug discovery because of their high speed and accuracy. Three classification models for A3 AR modulators were successfully built. Models from the training sets showed excellent results with the accuracy, sensitivity, and specificity values higher than 90%. AUC and MCC values were over 0.9. In addition, 3D-QSAR study was carried out to understand the quantitative relationship between the 3D-structure and affinity of the modulators resulted in quite reliable models. 3D-QSAR study for the agonists resulted in good cross-validated correlations obtained by CoMFA (q² of 0.594 and r² of 0.937) and CoMSIA (q2 of 0.560 and r2 of 0.907). The predictive ability of CoMFA and CoMSIA for the test set gave good predictive correlation coefficients of 0.768 and 0.730, respectively. For the antagonists, the CoMFA and CoMSIA analysis yielded significant cross-validated q² of 0.726 (r² = 0.913) and q² of 0.665 (r² = 0.915), respectively. The predictive correlation coefficients of CoMFA and CoMSIA in the test set were 0.808 and 0.767. Overall, in silico classification models and 3D-QSAR models were successfully built, and they confirmed that the presence of the hydrogen bond donor in 5’-amide position was a key criterion to distinguish agonists from antagonists for A3 AR. They could be useful not only to classify the agonists and antagonists but also to develop more potent A3 AR modulators.;Adenosine receptor (AR)는 G protein coupled receptor (GPCR)의 rhodopsin family에 속하고, A1, A2A, A2B, A3 네 개의 subtype으로 분류할 수 있다. 그 중 가장 최근에 밝혀진 A3 AR은 다양한 질병치료의 타겟 단백질로 생각되고 있다. A3 효능제는 허혈성 심질환과 허혈후 재관류시 조직손상(reperfusion injury) 등의 치료제로서 가능성을 가지고 있고, 길항제는 천식, 녹내장, 염증 치료제로 연구 중에 있다. 이 논문에서는 두 가지 리간드기반 접근 방법 (ligand-based approaches)을 적용한 실험을 하였다. 우선 Laplacian-modified naive Bayesian, recursive partitioning, support vector machine의 세 가지 machine learning 방법을 통해 분류기 모델(classification model)을 만들었다. 이 방법은 빠른 속도와 정확함을 장점으로 하여, 신약개발의 초기 단계에서 널리 사용되고 있다. 만들어진 모델을 여러 가지 수치를 이용하여 평가한 결과 정확성, 민감도, 특수성은 90% 이상, AUC와 MCC는 0.9 이상의 좋은 수치를 나타냈다. 다음으로, 효능제와 길항제 각각에 대해 3차원적인 정량적 구조-활성관계(three-dimensional quantitative structure-activity relationship; 3D-QSAR)에 대한 연구를 진행하여 신뢰성있는 model을 얻었다. 효능제의 경우 cross validation 했을 때, CoMFA의 q2이 0.594, r2이0.937의 값을, CoMSIA의 q2이0.560, r2이 0.907의 값을 나타냈으며 test set은 각각 0.768, 0.730의 r2을 보였다. 길항제에 대한 cross validation 결과 역시 CoMFA의 q2 이 0.726, r2 이 0.913을, CoMSIA의 q2이 0.665, r2이 0.915을 나타냈고, test set 예측에서는 각각의 r2이 0.808, 0.767의 높은 수치를 보였다. 분류기 모델과 3D-QSAR model 모두 성공적으로 만들어졌고, 이를 통해 agonists와 antagonists 분류에 5’-amide 위치의 수소 결합 donor의 존재유무가 중요한 요인이라는 것을 확인하였다. 이 두 모델은 앞으로 classification은 물론 새로운 효능제와 길항제 개발에 도움이 될 것으로 기대된다.-
dc.description.tableofcontentsI. INTRODUCTION = 1 II. METHODS = 4 A. Data set preparation and descriptor selection = 4 B. Machine learning methods = 5 1. Laplacian-modified nai¨ve Bayesian = 5 2. Recursive partitioning = 6 3. Support vector machine = 7 4. Classification model validation = 7 C. Molecular alignment for QSAR = 9 D. 3D-QSAR studies using CoMFA and CoMSIA = 9 III. RESULTS AND DISCUSSION = 11 A. Machine learning results = 18 B. 3D-QSAR results = 28 1. Statistical results of 3D-QSAR studies = 31 2. CoMFA and CoMSIA contour maps = 41 a. Contour maps of the agonists = 42 b. Contour maps of the antagonists = 44 IV. CONCLUSIONS = 46 V. REFERENCES = 48 국문초록 = 51-
dc.formatapplication/pdf-
dc.format.extent1166729 bytes-
dc.languageeng-
dc.publisher이화여자대학교 대학원-
dc.titleMachine Learning and 3D-QSAR Studies of A₃ Adenosine Receptor Modulators-
dc.typeMaster's Thesis-
dc.format.pageⅷ, 53 p.-
dc.identifier.thesisdegreeMaster-
dc.identifier.major대학원 생명·약학부약학전공-
dc.date.awarded2010. 2-
Appears in Collections:
일반대학원 > 생명·약학부 > Theses_Master
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

BROWSE