Reverse engineered gene networks reveal markers predicting the outcome of breast cancer

It is clear that gene-expression profiling measurements have great potential for improving breast cancer management and for increasing our understanding of the disease biology. Tens of works unveiled that there are many and discordant signatures with similar performances. In this work we propose a new approach to the breast cancer classification problem based on the adoption of whole organism gene network reverse engineering. We use publicly available dataset to reverse engineer a generic quantitative network for human. This network filter out the top ranked genes according to their network perturbation for each patient in a study cohort. The results are used as input of a classification algorithm. Although other studies used gene networks for classification purposes, none of them used reverse-engineering algorithms based on ordinary differential equations. We obtained performance indicators better than those reported in literature for what concern AUC and hazard ratio. We couldn't achieve the sensitivity needed for clinical use, but this work suggests that the approach is suitable for further investigation.