Flow chart to show the procedures of the method for inferring latent oncogenes. Seven protein-protein interaction (PPI) networks are constructed using the PPI information reported in STRING, from which feature vectors of oncogenes and unlabelled genes are extracted via Mashup. Unlabelled genes are divided into 39 parts; each of which combines the oncogenes to comprise a dataset. Each dataset induces a random forest (RF) model. A level value is computed for each unlabelled gene, which is the average of the probabilities yielded by 38 RF models. Unlabelled genes are ranked by the decreasing order of their level values. The same procedures are done for each oncogene, thereby ranking oncogenes. Finally, all genes are sorted in a list and ROC and PR curves are plotted to evaluate the performance of the method.