基于谷物蛋白质序列与PPI网络的功能预测研究

Function Prediction Based on Grain Protein Sequence and PPI Network

摘要:

谷物中现存大量未经注释、功能未知的蛋白质，且难以通过实验验证，因此计算方法成为预测谷物蛋白质功能的主流方法之一。作者以玉米、小麦、籼稻、粳稻4种谷物蛋白质为研究对象，利用数据库获取结构域相互作用信息。从蛋白质中较为稳定的结构域信息出发，结合AdaBoost算法获得蛋白质相互作用信息并构建蛋白质相互作用网络，将其与利用blast所获得的蛋白质序列相似性网络相结合，利用协同分类和多层感知机两种算法实现对谷物蛋白质的功能预测。研究结果显示，两种算法均能较为准确地预测蛋白质功能，其中协同分类在召回率方面表现更优，而多层感知机在准确率方面表现更优。本研究为谷物蛋白质的功能注释提供了新思路、新方法，对谷物的加工与营养研究提供了依据。

Abstract:

There are many unannotated proteins with unknown functions in cereals, which are difficult to be verified by experiments. However, computational methods have become one of the mainstream methods to evaluate the functions of the cereal proteins. In this study, maize, wheat, indica rice and japonica rice proteins were studied, and their structural domain interaction information was obtained from the related databases. The protein-protein interaction information was obtained and the protein-protein interaction network was constructed, starting from the relatively stable domain information of protein and combining the AdaBoost algorithm. Combining it with the biomolecular structure similarity network obtained by blast, the functions of the cereal proteins were predicted based on the cooperative classification and multi-layer perceptron algorithms. Results showed that both of cooperative classification algorithm and multi-layer perceptron algorithm could accurately predict the functions of the proteins. Moreover, collaborative classification algorithm showed a better recall rate, whereas multi-layer perceptron algorithm showed a better accuracy. This sudy revealed a new idea and method for the unannotated protein investigation, and provided a strong guarantee for cereal processing and nutrition research.