Scientific literature is the most reliable and comprehensive source of knowledge about molecular interaction networks. This knowledge is scattered in scientific literature written in natural languages, much time and labor have to be spent on manually extracting biological molecule interactions from literature. There have been many efforts for automatic extraction of biomedical knowledge from literatures. We propose a pattern matching algorithm with multiple Part-Of-Speech tagging based rules which could effectively reduce the required number of patterns and increase the recovery rate of traditional pattern matching algorithm. Various situations in biomedical texts are studied in the paper. The recovery and accuracy rate of the system is estimated to be 68.7% and 93.0%, respectively.