Motivation: Like most human diseases, tumors are complex traits, the genesis and development of which recruit a number of genes and several important biological processes. As proteins involved in common processes tend to be centralized in the same local area of protein-protein interaction networks, here a novel framework has been developed to identify which areas of the networks are most relevant to a phenotype.
Results: These areas termed 'coherent modules' can be regarded as gene sets dynamically defined in the networks. Compared with previous analogous approaches, one critical feature of our method is the optimization of coherent modules for two distinct aspects balanced by tuning a parameter in the framework. First, we seek the low coupling between coherent modules and then maximize the intrinsic similarity within a module. The framework has good expansibility, with classical expression data analysis methods generalized as particular cases. This coherent module approach was applied to an oral cavity tumor dataset with 18 significant coherent modules identified, which could indicate the presence of lymph node metastasis. Further examination shows that most of the modules are responsible for comparatively independent biological processes. Our framework is helpful for the prognosis of tumors and offers a new perspective for tumor research.
Supplementary information: Supplementary data are available at Bioinformatics online.