The initiation and progression of complex diseases have a close relationship with dysfunction of biological pathways in our body. Developing computational techniques to study the relationship between diseases and pathways through high-throughput data has essential biological significance. However, the traditional identification approaches of pathways which are significantly related to experiment conditions usually reduce pathways to gene sets. It is obvious that these methods do not consider the interactions between genes and the different roles that genes play in pathways, and they don't fully mine pathway information. Therefore we integrated protein-protein interaction information and gene weights into pathway analysis, and constructed a protein-pathway network which contains information in protein-protein interactions and pathways. We then scored pathways by random walk algorithm to optimize disease risk pathways. Finally, the statistically significant pathway can be identified through permutation method. We applied the network to a colorectal cancer dataset, and finally identified fifteen pathways which are significantly related to the development of this disease. Compared with other pathway identification methods (hypergeometric test and SPIA), our approach can effectively identify risk pathways related to complex diseases. In order to test the stability of our method in identifying risk pathways related to diseases, we used our method to identify risk pathways by using another colorectal cancer dataset. We found that the identified results can prove the stability of our method.