Implementation of SET.(A)(see Additional file 1) 10 metastatic tumours and 13 non-metastatic tumours were assigned into M and P group, respectively. (B)(see Additional file 2) Error rate distribution of expression signature genes. This plot suggested the top 3~4 genes are capable of being the best signature to distinguish samples (error rate 0.087). (C)(see Additional file 3) "User-defining" interface allows user to select/de-select features. (D)(see Additional file 4) The plot shows the prediction strength of a user-defined signature (top 4 genes from 2B) in discriminating non-metastatic and metastatic tumours. The table above PS plot shows the significance of the error rate (less than 0.001 in this case). (A) – (D) show the relevant sections of the original software interface. For the full images please see Additional files 1-4, respectively.