Identification of functional modules using distributed graph clustering for mechanism-based diagnosis of cancer

Abstract

Various method for discovering disease-related genes have been developed for the understanding of disease mechanism, diagnosis, and therapy. As the knowledge on cellular functions is growing, the research area is expanding to search for the disease-related functions and its underlying mechanisms. Nevertheless, various information of functional modules is unorganized, and there was a lack of research on discovery or reconstitution of basis mechanisms by integrating mechanism information and omics-analysis results. In this study, a method to reconstitute novel modules was developed using agglomerative clustering with the integration of several types of relations of genes within modules. Moreover, a module discovery method, named DFENS-G, was developed. DFENS-G scores reconstituted modules by integrating similarity with known cancer markers and the significance of activation score of modules in Omics. In comparison with the previous marker discovery method, the model constructed with selected modules from DFENS-G shows the best accuracy of prediction in liver, colorectal, and breast cancer. Besides, DFENS-M was applied to predict the response of Sorafenib, a drug for liver cancer, and showed that it could successfully discover the module markers which can be used in companion diagnosis. As a result, DFENS-M system can improve the confidence of the markers by integrating previous mechanism information and propose an explainable model.