Gil Perez, G. J.; Perez Rodriguez, R.; Gonzalez, A.

Background: The complexity of gene regulatory networks, involving thousands of genes, poses a fundamental challenge to understanding cancer phenotype reversal. However, recent evidence suggests that the effective dimensionality of normal and tumor transcriptional manifolds is low, and that small panels of genes can discriminate perfectly between normal and tumor samples. Methods: We build upon two previously developed concepts: (i) highly accurate normal and tumor gene markers (namely, N- and T-markers), defined as genes with exclusive expression intervals in normal and tumor samples, respectively; and (ii) gene deregulation networks (GDNs), represented as directed acyclic graphs encoding causal relationships between gene deregulation events. A subset of genes appearing in both marker classes (NT-markers) act as bridging nodes between the N- and T-GDNs. Starting from these elements, we introduce a quantitative dynamical model based on node frequency and connectivity to assess how gene intervention effects propagate through the GDN and thereby predict their overall impact on the tumor tissue. Results: According to the model, interventions on pure T-markers (T-markers that are not NT-markers) produce effects largely confined to the T-GDN, with a minimal perturbation of the N-network. Interventions on pure N-markers (N-markers that are not NT-markers) generate a perturbation of both networks, but with limited effect. In contrast, interventions on NT-markers with high activation frequency in both tumors and normal state (e.g., AGER in lung adenocarcinoma: 98% in tumor samples, 75% in normal samples) can induce bidirectional phenotype shifts. For an effective combination of targets, coverage across tumor samples must be maximized. At the same time, in the T-GDN the number of nodes unreached by the reverse cascade following the intervention must be minimized, as these regions may act as escape routes for the tumor. Escape probability further depends on the tumor stage and the tumor activation rate of new T-genes. When targeting NT genes, high frequency in normal samples should also be prioritized. Conclusions: High-frequency NT-genes, due to dual network connectivity and tissue relevance, represent optimal targets for achieving at least partial phenotype reversal. This framework provides a quantitative guide for prioritizing gene therapy targets and designing combination strategies that balance coverage, escape minimization, and normal tissue relevance.