| Summary: | 碩士 === 國立中興大學 === 基因體暨生物資訊學研究所 === 103 === Glycosylation is a significant and common post-translational modification of proteins, which plays important roles in various biological processes. The protein folding, activity, structure, stability and other biological properties are affected by glycosylation, which is closed with Alzheimer’s disease and cancer. Therefore, this study designed a new comprehensive information model to develop a prediction tool GlycoCI, including 4 kinds of single models, for predicting N-, O- and C-linked glycosylation sites, oreover, and the bootstrap sampling method was compared to comprehensive information model for the effectiveness validation. For the selection of feature combination, we considered six kinds of features, including sequence, structure, and functional based features. On the other hand, 68 different kinds of classifiers were evaluated for improving the performance of GlycoCI and the RandomCommittee, MultilayerPerceptron, Bagging and FilteredClassifier are the best models corresponding to four single models. Two independent sets were used for evaluation and comparison between GlycoCI and other glycosylation site prediction tools. Finally, we further analyze the performance of GlycoCI on different species and protein subcellular localization. GlycoCI provides more accurate prediction on protein glycosylation prediction and a web server was constructed with freely available at http://predictor.nchu.edu.tw/GlycoCI.
|
|---|
