Figure 1.

Screening-network procedure. (a) BMD-screening, DAD-screening or BMD+DAD-screening are applied to the training dataset T to choose a subset of genes $\left\{x_j,j\in \mathcal{I}\right\}$. (b) We use subset $\left\{x_j,j\in \mathcal{I}\right\}$ to incorporate the biological information as networks of interactions. (c) Network-penalized methods are executed on $\left\{x_j,j\in \mathcal{I}\right\}$. (d) High-risk genes or potential biomarkers (${\widehat{\beta }}_{\mathcal{I}}\ne 0$) are selected (from each screening approach) and are used to separate patients in two groups (high-and-low risk group). (e–f) The performance of the survival prediction are assessed using a testing dataset D. To make sense of the gene signatures (g) Pathway analysis is performed on high-risk genes and (h) COSMIC investigation is carried out.