Protein-protein interaction (PPI) is important in the field of experimental biology as well as bioinformatics. In BioCreative V, we participated in the BioC task and developed a PPI system to detect pas-sages with PPIs in the full-text articles. By adopting the BioC format, the output of the system could be seamlessly added to the biocuration tool with little effort required for the system integration. Our PPI system utilizes Extended Dependency Graph as an intermediate level of representation to abstract away syntactic variations in the sentence. Asa result, we only use three basic predicate-argument rules to extract PPIpairs in the sentences, and two additional rules to detect additional pas-sages with PPI pairs. Experiments on 20 in-house full-text articles show that we are able to obtain a recall of 77.8. By using only three basic rules, experiments on AIMed further confirm that we can achieve a precision of 91.5 of sentence selection and an F-value of 62.8 of instance selection.