Figure 4

Random prognostic gene signatures for PDAC. (A–E) Randomisation results of 5 million gene signatures with the initial feature set of 225 genes. All signatures were trained and validated on the same datasets as used for the derivation of the 36-gene signature. A χ² statistic was used a performance metric for comparing random signatures with the 36-gene signature. The dashed lines are the χ² values for the 36-gene signature. Overall, only 0.19% of the random signatures outperformed the 36-gene signature across all datasets. (F) Percentage inclusion of genes in the 36-gene signature for the randomly selected gene signatures. Thirty-one out of 36 genes were amongst the top 5% genes with highest inclusion frequency, with SEMA3A ranked highest (overall ranked 11th). PDAC, pancreatic ductal adenocarcinoma.