Improving Protein-Protein Interaction Sites Prediction by Using Structure-Based Profiles

• Main Authors: I-pin Su, 蘇義斌
• Other Authors: Chin-sheng Yu
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/27492915379999549685

碩士 === 逢甲大學 === 資訊工程所 === 99 === Protein-protein interaction sites occur in residue-residue contact surface on the interacting proteins. Identifying the interaction sites assists in understanding the function of the protein complex. Here, we proposed a method that exploited support vector machine algorithm to identify protein-protein interaction sites from structure-based profiles. A non-homologous dataset of 333 protein complexes with 1134 chains was used for training and testing evaluation. Not only the sequence evolutionary profile (PSSM), we also used multiple structure-based profiles as input features. The profiles are the scoring matrices of the statistical probability distribution, such as secondary structure and solvent-accessible surface area. The results reveal that using multiple structure-based profiles will promote the predicted accuracy, and also provide the probability of keep on improving the methods in the future.