Cancer diseases could possibly induced by abnormal protein-protein interactions (PPIs). Given the PPIs (Biogrid), cancer genes (Tumor Associate Gene, TAG) and domain annotation (InterPro) information, we derived the onco-proteins (OCP) and tumor suppressor proteins (TSP) domain-domain interactions (DDIs). Two interaction models are introduced to describe cancerous proteins domain-domain interactions (DDI), that is the single domain interaction model and the multi-domain interaction model. The cancer-specific DDIs scoring information are employed to valid the single domain interaction model for the OCP-OCP, TSP-TSP and OCP-TSP dataset. It is found that the OCP-OCP DDIs score achieved the best sensitivity test result, which is 79.9%. The TSP-TSP DDIs score obtained the lowest sensitivity result, which is 54.8%. Furthermore, the lowest false positive is obtained by training the TSP-TSP DDIs score against the OCP-OCP test set, which has a 11.6% false positive rate. On the other hand, we also compared the top 20 OCP-OCP DDIs for both the single domain interaction model and multi-domain interaction model, it is found that 60% of the DDIs overlapped, hence, the multi-domain interaction model could provide much more DDIs pairs information, but it does not incur much effects on their ranking. A web-based interface has been set up to report our results, which is available at http://210.70.83.81/tsgocg-ppi/.
