Abstract: With the development of proteomics,scholars begin to pay more attention to the construction of Protein-Protein Interaction (PPI) network.Mass spectrometry(MS) has become a representative method for protein-protein inte-raction (PPI) inference,and it is one of the main experiment method to construct PPI network.Based on the technology of mass spectrometry,a large amount of experimental protein MS data is generated,such as affinity purification-mass spectrometry (AP-MS) data and protein correlation profiling-mass spectrometry (PCP-MS) data,which provide important data support for the construction of PPI networks,but constructing PPI networks by hand is impracticable and time consuming.Thus,PPI network inference algorithm for PCP-MS data has begun to become the research hotspot in bioinformatics.This thesis focused on the problem of PPI network inference for two main types of mass spectral data (AP-MS data and PCP-MS data),and designed effective methods respectively to solve the issue of current bottlenecks,achieving the construction of high-quality PPI network.The existing algorithms for PPI network interface from PCP-MS data are still in infancy,and there is a few of related algorithms.The existing method have several problem.Specifically:1)many error interaction is contained in the results produced by the different algorithms,and the correct interaction is omitted in the results.2)Different algorithms may produce very different results when they face the same data set.3)For different data sets,the performance variance of the same algorithm is larger.For the problem of PPI network inference for PCP-MS data,this paper proposed a PPI scoring method based on correlation analysis and rank aggregation.The method is based on unsupervised learning and includes two steps.Firstly,correlation coefficient between protein pairs is computed,and multiple results of PPI scores can be obtained.Secondly,multiple results for each pair of proteins are combininect via rank aggregation to a single PPI score.The experimental results show that this method is comparable with those supervised learning methods using standard reference set.

Key words: Correlation analysis, MS data, PPI network, Protein direct interaction, Rank aggregation