N-Gene and T-Gene deregulation networks: a data-driven causal framework for the analysis of gene interventions in...

Current gene regulatory networks are limited by incomplete functional annotation and the difficulty of inferring causal relationships from expression data. Here we introduce Gene Deregulation Networks (GDNs), a new structure in which a directed link from gene C to gene E indicates that a deregulation of C makes a deregulation of E more probable. GDNs are inferred from expression data using a probabilistic theory of causation, without requiring prior biological knowledge.

Using previously defined N- and T-genes with exclusive expression intervals for normal tissue and tumors, respectively, we construct separate GDNs for normal and for tumor tissues. Data are TCGA RNA-Seq bulk profiles from five cancer types. Links are identified via the Loevinger coefficient and pruned with Reichenbach-type and Mokken tests. We then project each sample onto its corresponding Gene Deregulation Network to visualize the deregulation cascades that have occurred. Finally, we define a simple dynamics: spontaneous evolution follows the direction of the GDN edges; interventions (e.g., gene knockdown) acting against spontaneous evolution induce cascades along the reversed network.

The GDNs are represented by sparse, directed acyclic graphs. Genes with low deregulation frequency have high out-degrees, suggesting they act as upstream regulators. High-frequency genes have high in-degrees, indicating they are convergence points of cascades. Projecting samples onto the T-GDN reveals that early tumors rely mostly on spontaneous T-gene activations, whereas advanced tumors show wide, branching cascades. The N-GDNs show consistent size and structure across distinct tissues revealing similar protective machinery against tumor formation. In contrast, the T-GDNs quantitatively differ from tissue to tissue indicating different levels of transcriptional reprogramming. Simulated knockdown of EPHA10 and of an 8-gene panel illustrates how the network topology determines whether an intervention can be resisted by the tumor. A reported experiment on POM121 knockdown in two prostate cancer cell lines qualitatively confirms the predicted directionality of cascades.

GDNs provide a robust, scalable, and annotation-free framework to understand cancer onset and progression. The separation into N- and T-GDNs, connected by NT-genes, offers a systematic basis for studying carcinogenesis and designing targeted therapies.