歐陽彥正Oyang, Yen-Jen臺灣大學:資訊工程學研究所朱文藝Chu, Wen-YiWen-YiChu2010-05-182018-07-052010-05-182018-07-052008U0001-2407200817325300http://ntur.lib.ntu.edu.tw//handle/246246/183600本篇論文旨在設計一個能從多肽序列擷取資訊的自動分類器，以預測轉錄因子上會與DNA之鹼基產生鍵結的殘基。正如最近一些研究所揭露的，有大量轉錄因子之三級結構是不穩定序(disordered)，因此若能只純粹利用序列的資訊，進而預測轉錄因子與DNA之關鍵殘基，將非常有助於下一步的實驗。鑑於此，設計、發展一個預測器並使之能夠分辨與DNA進行專一性結合的殘基更形迫切需要。此外，專一性結合不僅能反應基因上的特異序列辨識，在正確的基因調控中也扮演極重要的角色。論文呈現的設計混合了兩種不同方法，分別是以SVM為主的機器學習方式以及原先應用於預測蛋白質域(protein domain)的演算法。觀察後發現兩個方法的預測表現在不同的蛋白質二級結構上各有優劣，於是我們嘗試設計一套機制以混合兩種方法的輸出結果以取得最佳的成績。在本文最終的實驗結果，呈現的新預測器能提供59.5%的涵蓋度、77.4%的精確度，以及98.8%的專一度(specificity)。This thesis presents the design of a polypeptide sequence based predictor aiming to identify the residues in a transcription factor that are involved in specific binding with the DNA. As a recent study has revealed that the tertiary structures of a large number of transcription factors are mostly disordered, the capability to identify the residues in a transcription factor that play key roles in interaction with the DNA based purely on analysis of the polypeptide sequence is highly desirable. In this respect, it is further desirable to have a predictor capable of distinguishing those residues involved in specific binding with the DNA, since specific binding corresponds to sequence-specific recognition of a gene, which is essential for correct gene regulation. The design of the proposed predictor is distinctive by employing a hybrid approach. That is, two prediction mechanisms specialized for making predictions in different types of protein secondary structures have been incorporated. In the experiments reported in this thesis, the proposed hybrid predictor delivered precision of 77.4%, sensitivity of 59.5%, and specificity of 98.8%Contents 謝 I 要 IIBSTRACT IIIhapter 1 INTRODUCTION 1hapter 2 RELATED WORK 7.1 OVERVIEW 7.2 CLASSIFIER 10.3 FEATURE SET 15hapter 3 THE PROPOSED HYBRID PREDICTOR 17.1 OVERVIEW 18.2 PRIMARY SVM PREDICTOR 20.3 AUXILIARY SSEP PREDICTOR 21hapter 4 EXPERIMENTAL RESULTS 25.1 DESIGN OF EXPERIMENTS 25.2 RESULTS AND DISCUSSIONS 26hapter 5 CONCLUSIONS AND FUTURE WORKS 36EFERENCES 38PPENDIX 43application/pdf773872 bytesapplication/pdfen-US專一性結合殘基轉錄因子secific-binding residuetranscription factorthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/183600/1/ntu-97-R95922048-1.pdf