Context-Dependent Networks
Context-dependent networks are relational systems in which the existence, strength, or functional meaning of connections between components varies with the state of the system itself. Unlike the static graphs of classical network topology, where edges are fixed properties of the underlying structure, context-dependent networks treat connectivity as a dynamical variable: who is connected to whom, and with what consequence, depends on where the system currently sits in its state space. This is not a complication to be engineered away. It is the defining feature of biological, neural, and social systems.
The concept arises whenever a network's edges are modulated by the activity levels of the nodes they connect. In a protein-protein interaction network, the "edge" between two proteins is not a permanent bond but a conditional event: it occurs when concentrations exceed thresholds, when post-translational modifications expose binding domains, when compartmentalization brings reactants into contact. The network topology measured in a yeast two-hybrid assay is not the topology of the living cell. It is a population average of context-dependent interactions that may never occur simultaneously in any single cell.
From Structure to Dynamics
The formal shift from static to context-dependent networks requires replacing the adjacency matrix — a binary or weighted matrix that summarizes all edges — with a state-dependent coupling function. In a context-dependent framework, the coupling between node i and node j is not a constant A_{ij} but a function A_{ij}(x,t) of the system's state vector x and possibly time t. This seemingly minor change has profound consequences. The spectral properties that determine network synchronizability, the centrality measures that identify influential nodes, and the community structures that partition the network into modules all become transient properties that must be recomputed as the system evolves.
This dynamic reframing reveals why classical graph-theoretic analysis systematically misleads when applied to regulatory systems. A node with high betweenness centrality in the static graph may be dynamically inert — its position is structurally prominent but its regulatory influence is buffered by saturating kinetics or competing interactions. Conversely, a node with low static degree may act as a dynamical switch: when activated, it reconfigures the effective network topology by recruiting or suppressing downstream pathways. Network pharmacology has repeatedly failed to predict such switches because it searches for structural hubs rather than dynamical thresholds.
Context in Biological Systems
Biological regulatory networks are the paradigmatic case of context-dependent structure. A gene regulatory network is not a wiring diagram but a repertoire of possible wirings, each expressed under different cellular conditions. The same transcription factor can activate one target in a stressed cell and repress it in a resting cell, depending on cofactor availability, chromatin state, and metabolic signals. The "network" is not the set of possible edges but the set of edge-activation rules — a higher-order structure that static topology cannot represent.
This has direct implications for drug discovery. A drug that targets a protein hub may be effective in one cellular context and irrelevant in another, not because the target is absent but because the target's regulatory role is context-dependent. The failure of many network pharmacology predictions stems from treating context as noise to be averaged away rather than as the signal that determines therapeutic relevance. The future of systems pharmacology lies not in bigger graphs but in state-dependent coupling models that track how the effective network rewires under disease progression and therapeutic intervention.
Context in Social and Information Systems
Social networks are equally context-dependent, though the relevant "state" is often harder to specify. The influence of an interpersonal tie varies with situational factors: the same friendship that enables information flow during a crisis may suppress dissent during routine periods. The strength of a link is not a property of the dyad but a property of the dyad embedded in a situation. Social network analysis that treats tie strength as a constant attribute of the relationship systematically underestimates the volatility of influence propagation.
Information systems exhibit context-dependence through algorithmic curation. The recommendation network that connects users to content is not a static graph but a dynamically rewired structure driven by engagement metrics, user history, and real-time feedback. The "edges" of the content graph are reweighted every millisecond, creating a system whose topology is inseparable from its dynamics. This is why temporal graph theory — the study of networks whose edges have explicit time-varying weights and durations — is becoming essential for understanding online platforms.
The insistence on context-dependence is not a call for more data. It is a call for a different mathematics. The graph is the wrong abstraction for systems whose connectivity is itself a dynamical variable. Until network science abandons its dependence on static topology and embraces the full state-dependence of real systems, it will remain a discipline that computes beautiful properties of graphs that do not exist.