Support vector machines to predict DNA-binding proteins

• Main Author: 劉家輝
• Other Authors: 吳哲賢
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/89198735918439732887

碩士 === 中華大學 === 資訊工程學系碩士班 === 92 === In this work, we have developed a prototype tool for binary prediction of DNA-binding proteins. We combine segmentation of protein sequences using Poisson generator and extraction of distance features which are used as input for classification by support vector machine (SVM) which is first shaped by the training set to predict the DNA-binding proteins. To extract the distance features, 40-dimensional vectors together with the limited range correlation of charge, hydrophobic, and accessible surface area are used as input to represent the amino acid composition, and each of the tools predicts whether the protein belongs to DNA-binding proteins or Non-binding proteins. The measures have facilitated the success of learning and prediction that the results have consistently achieved approximate 78.4% accuracy.