Protein-protein interaction is an important issue in proteomics research. In this study, two predition methods, including support vector machine (SVM) and information analysis of physical and chemical properties and sequence information of amino acid, were used to construct the feature vectors for predicting protein-protein interaction. Data of protein expression by 34000 pairs of yeast were taken from DIP database. The prediction accuracy by the method of sequence information of amino acid was 79.63%, and the prediction accuracy by the method of information analysis of physical and chemical properties 75.86%. An improved method was proposed by combing the above two to make the prediction accuracy up to 83.72%. Furthermore, the function of degree of disagreement (FDOD) was used to reduce the dimension of support vector.%蛋白质与蛋白质相互作用研究是蛋白质组学的重要研究内容之一.本研究采用支持向量机学习方法,将氨基酸物理化学特性和序列信息方法相结合构建支持向量,选取DIP数据库中的酵母表达蛋白序列进行蛋白质相互作用预测.在34 000对酵母表达蛋白实验数据中,预测准确率达到83.72%,而单独运用基于氨基酸物理化学特性的方法和基于序列信息的方法预测准确率分别为75.86%和79.63%.在提高预测准确率的同时通过引入离散信息度量函数(FDOD)减少支持向量的维数,使支持向量学习时间缩短,提高相互作用预测的速度.
展开▼
