Fig. 1.

iRafNet schematics. For each gene $g_{\text{j}}\in \{1,\mathrm{...},p\}$, we determine a ranked list of potential regulators via iRafNet. Based on each data $d\in \{1,\mathrm{...},D\}$, we derive weights $\left\{w_{k\to j}^{\text{d}}\right\}$ measuring the prior belief of regulatory relationships $\{g_{\text{k}}\to g_{\text{j}}\}$. Using expression data, we run random forest to find genes regulating g_j. At each node, instead of sampling a random subset of genes from the entire set of genes; we randomly choose an integer $I\in \{1,\mathrm{...},D\}$ and we sample genes according to weights $\left\{w_{k\to j}^{\text{I}}\right\}$. The final network is derived by ranking potential regulators based on the random forest importance score