基于Minimum Cuts的蛋白质交互识别

摘要

Protein-Protein Interaction (PPI) information is significant for biological and medical research,and is an important content in biomedicine field.The recognition of PPI with large-scale corpus can significantly reduce the cost of manual annotation by directly using the existing PPI database.Therefore,a method for PPI with Minimum Cuts based on the large-scale corpus has been proposed.Based on the framework of relational similarity,Minimum Cuts classifier not only uses SVM to predict the classification initially of a single protein,but also makes use of the similarity between the protein pairs to determine the results which are more accurate.The experimental results show that the Minimum Cuts classifier is better than the SVM classifier for the recognition of PPI.When the training data is 20%,the recognition results of the Minimum Cuts classifier gets better performance than that of an SVM classifier trained with 80%.%蛋白质交互信息对生物、医药研究有着重要意义,是生物医学领域一项重要的研究内容.对基于大规模语料库的蛋白质交互识别,直接利用已有的PPI数据库,能显著降低人工标注的代价.为此,在大规模语料库的基础上,提出了基于Minimum Cuts的蛋白质交互识别方法.在关系相似性框架下,Minimum Cuts分类器不仅采用SVM算法对单个蛋白质对进行初步分类预测,还利用蛋白质对之间的相似性约束判断结果,使分类结果更加准确.实验结果表明,利用Minimum Cuts分类器进行PPI的识别结果优于SVM分类器的识别结果.当训练数据为20%时,Minimum Cuts分类器的识别结果优于训练数据为80%时的SVM分类器的识别结果.