HCC prognostic signatures are not unique

Throughout the past decade, multiple molecular signatures predicting the course of Hepatocellular carcinoma (HCC) were reported. However, these signatures have not yet been integrated into clinical routine decision making. Investigating randomly chosen signatures of various size, we demonstrate that approximately 15.7 to 30.1% of all random signatures exhibit prognostic potential. This finding was further confirmed by using predefined, functionally related gene sets such as GSEA, KEGG pathways or GO terms.

Given the large number of potentially prognostic random signatures, we further investigated whether combining multiple signatures would result in an improved significance and stability of prognostic capability. These combined analyses demonstrated that HCC patient samples may be divided into high predictable and low predictable samples. In high predictable samples most random signatures resulted in comparable prognosis.

In turn, we demonstrate by training and independent validation, that if most random signatures show the same prognosis, this result must be considered highly reliable. Thus, using “swarm intelligence” for prognostic evaluation of HCC may be superior to single prognostic signatures of any size and increase stability of prognosis.